I had same issues and I didn't take separate regulator too seriously, which was mistake.
So I made it work using
a. separate voltage regulator
b. capacitor cca 2uF between + and - pins on NRF24
c. using RF24_PA_MAX
d. another device was about 2 meters away

Before that I had only capacitor, and when I put voltage regulator it worked for me.
I took power from arduino nano (+5V pin). No need for separate power. Powering Arduino Nano over USB is fine.

I have similar solution so I can tell
a) Flickering is not visible at all. Not even close to low frequency monitors we had before LCD come (maybe you remember that times)
b ) lifetime of led should not be afected, at least could be ignored (if you will replace LED after 5 or 7 years ....it doesn't matter much, since it will last much more than traditional bulbs)
c) Power consumption is current times voltage. Voltage is constant, and current if at 100% is for maximum brigtness. If you use arduino PWM then 50% means that at 50% you have current, and vice versa. Effectively it means half the current. And that means half the power. So your assumption is correct.
MOSFET has incredibly low resistance, so almost nothing is spent on MOSFET (comparing with traditional transistors as your question might imply)
d) Input power must cover your LEDs. If you use 50W LED than you should have at least 50W/12V = 4,16 A for maximum load. With some reserve i would probably take at least 5A supply. Real answer is more - how long you will really have 100% power.
In my project I usually pick some color so even if red is 100, and blue is 100, maybe green is 0%, or .... So in my case average is ~50% of maximum calculated power most of the time.
TO be on safe side pick 8A 12V supply (conservative) or take 5A 12V supply which in my case is more than enough because I almost never use white collor (all collors at maximum).

1. MOSFET cooling. It is incredibly low heat on MOSFET. In fact my power converter 12V to 5V for electronic is hot, while MOSFETS are incredibly cold. Again the fact that internal restistance is very low, helps a lot. I created a board with place for coolers. and I bought passive coolers (similar to these on link you provided) . But after testing I decided that that makes no sense. So I didn't put them on!

I am talking about my project expirience, where I use 5m 5050 RGB LED, with 8266 microcontroller and 3 MOSFETs - 1 for each color. I use 5A 12V power suply, and swithching module for adapting power for my 8266 microcontroler (otherwise power converter overheats).

I hope my expirience helps.

I have similar solution so I can tell
a) Flickering is not visible at all. Not even close to low frequency monitors we had before LCD come (maybe you remember that times)
b ) lifetime of led should not be afected, at least could be ignored (if you will replace LED after 5 or 7 years ....it doesn't matter much, since it will last much more than traditional bulbs)
c) Power consumption is current times voltage. Voltage is constant, and current if at 100% is for maximum brigtness. If you use arduino PWM then 50% means that at 50% you have current, and vice versa. Effectively it means half the current. And that means half the power. So your assumption is correct.
MOSFET has incredibly low resistance, so almost nothing is spent on MOSFET (comparing with traditional transistors as your question might imply)
d) Input power must cover your LEDs. If you use 50W LED than you should have at least 50W/12V = 4,16 A for maximum load. With some reserve i would probably take at least 5A supply. Real answer is more - how long you will really have 100% power.
In my project I usually pick some color so even if red is 100, and blue is 100, maybe green is 0%, or .... So in my case average is ~50% of maximum calculated power most of the time.
TO be on safe side pick 8A 12V supply (conservative) or take 5A 12V supply which in my case is more than enough because I almost never use white collor (all collors at maximum).

1. MOSFET cooling. It is incredibly low heat on MOSFET. In fact my power converter 12V to 5V for electronic is hot, while MOSFETS are incredibly cold. Again the fact that internal restistance is very low, helps a lot. I created a board with place for coolers. and I bought passive coolers (similar to these on link you provided) . But after testing I decided that that makes no sense. So I didn't put them on!

I am talking about my project expirience, where I use 5m 5050 RGB LED, with 8266 microcontroller and 3 MOSFETs - 1 for each color. I use 5A 12V power suply, and swithching module for adapting power for my 8266 microcontroler (otherwise power converter overheats).

I hope my expirience helps.

I had same issues and I didn't take separate regulator too seriously, which was mistake.
So I made it work using
a. separate voltage regulator
b. capacitor cca 2uF between + and - pins on NRF24
c. using RF24_PA_MAX
d. another device was about 2 meters away

Before that I had only capacitor, and when I put voltage regulator it worked for me.
I took power from arduino nano (+5V pin). No need for separate power. Powering Arduino Nano over USB is fine.

@Igor-Antolić huh,... my mistake, just installed 2.1.0 instead of 2.0.0. It seems it is from 2017-03-07.

I dont have a good way to find last version of lib. Any suggestion how to check if last stabile version of mysensors java lib?

Hi, did anyone hit the problem when it seems that all works fine
and then mysensors plugin start to think that it failed connecting to bridge (even if it doesnt actually)

Should I find newer version (last jar I found is since 2017-01-18)

I put message from node every 1 second, and in log you will see "0;255;3;0;9;TSF:MSG:READ...." every second
but every 10 second it recconects saying "Failed connecting to bridge...next retry in 10 seconds (Retry No.:31).... Shutting down serial connection!...

17:40:21.656 [DEBUG] [rs.internal.protocol.MySensorsReader] - 0;255;3;0;9;TSF:MSG:READ,202-202-0,s=1,c=1,t=37,pt=2,l=2,sg=0:468
17:40:21.657 [DEBUG] [rs.internal.protocol.MySensorsReader] - 202;1;1;0;37;468
17:40:22.421 [ERROR] [col.serial.MySensorsSerialConnection] - Can't connect to serial port. Wrong port?
17:40:22.424 [ERROR] [col.serial.MySensorsSerialConnection] - Failed connecting to bridge...next retry in 10 seconds (Retry No.:31)
17:40:22.425 [DEBUG] [col.serial.MySensorsSerialConnection] - Shutting down serial connection!
17:40:22.842 [DEBUG] [rs.internal.protocol.MySensorsReader] - 0;255;3;0;9;TSF:MSG:READ,202-202-0,s=1,c=1,t=37,pt=2,l=2,sg=0:486
17:40:22.844 [DEBUG] [rs.internal.protocol.MySensorsReader] - 202;1;1;0;37;486
17:40:24.029 [DEBUG] [rs.internal.protocol.MySensorsReader] - 0;255;3;0;9;TSF:MSG:READ,202-202-0,s=1,c=1,t=37,pt=2,l=2,sg=0:470
17:40:24.032 [DEBUG] [rs.internal.protocol.MySensorsReader] - 202;1;1;0;37;470
17:40:25.218 [DEBUG] [rs.internal.protocol.MySensorsReader] - 0;255;3;0;9;TSF:MSG:READ,202-202-0,s=1,c=1,t=37,pt=2,l=2,sg=0:474
17:40:25.222 [DEBUG] [rs.internal.protocol.MySensorsReader] - 202;1;1;0;37;474
17:40:26.406 [DEBUG] [rs.internal.protocol.MySensorsReader] - 0;255;3;0;9;TSF:MSG:READ,202-202-0,s=1,c=1,t=37,pt=2,l=2,sg=0:486
17:40:26.407 [DEBUG] [rs.internal.protocol.MySensorsReader] - 202;1;1;0;37;486
17:40:27.591 [DEBUG] [rs.internal.protocol.MySensorsReader] - 0;255;3;0;9;TSF:MSG:READ,202-202-0,s=1,c=1,t=37,pt=2,l=2,sg=0:466
17:40:27.592 [DEBUG] [rs.internal.protocol.MySensorsReader] - 202;1;1;0;37;466
17:40:28.783 [DEBUG] [rs.internal.protocol.MySensorsReader] - 0;255;3;0;9;TSF:MSG:READ,202-202-0,s=1,c=1,t=37,pt=2,l=2,sg=0:483
17:40:28.786 [DEBUG] [rs.internal.protocol.MySensorsReader] - 202;1;1;0;37;483
17:40:29.967 [DEBUG] [rs.internal.protocol.MySensorsReader] - 0;255;3;0;9;TSF:MSG:READ,202-202-0,s=1,c=1,t=37,pt=2,l=2,sg=0:483
17:40:29.972 [DEBUG] [rs.internal.protocol.MySensorsReader] - 202;1;1;0;37;483
17:40:31.158 [DEBUG] [rs.internal.protocol.MySensorsReader] - 0;255;3;0;9;TSF:MSG:READ,202-202-0,s=1,c=1,t=37,pt=2,l=2,sg=0:463
17:40:31.163 [DEBUG] [rs.internal.protocol.MySensorsReader] - 202;1;1;0;37;463
17:40:32.344 [DEBUG] [rs.internal.protocol.MySensorsReader] - 0;255;3;0;9;TSF:MSG:READ,202-202-0,s=1,c=1,t=37,pt=2,l=2,sg=0:487
17:40:32.346 [DEBUG] [rs.internal.protocol.MySensorsReader] - 202;1;1;0;37;487
17:40:32.431 [DEBUG] [col.serial.MySensorsSerialConnection] - Connecting to /dev/ttyUSB0 [baudRate:115200]
17:40:32.435 [DEBUG] [col.serial.MySensorsSerialConnection] - Final port list: /dev/ttyUSB0
17:40:33.530 [DEBUG] [rs.internal.protocol.MySensorsReader] - 0;255;3;0;9;TSF:MSG:READ,202-202-0,s=1,c=1,t=37,pt=2,l=2,sg=0:470
17:40:33.533 [DEBUG] [rs.internal.protocol.MySensorsReader] - 202;1;1;0;37;470
17:40:34.438 [ERROR] [col.serial.MySensorsSerialConnection] - Can't connect to serial port. Wrong port?
17:40:34.440 [ERROR] [col.serial.MySensorsSerialConnection] - Failed connecting to bridge...next retry in 10 seconds (Retry No.:32)
17:40:34.441 [DEBUG] [col.serial.MySensorsSerialConnection] - Shutting down serial connection!
17:40:34.720 [DEBUG] [rs.internal.protocol.MySensorsReader] - 0;255;3;0;9;TSF:MSG:READ,202-202-0,s=1,c=1,t=37,pt=2,l=2,sg=0:472

@HenryWhite That is nice to hear that you ordered 2 items, however now I am confused because you said before that you thik aboit "E01-ML01DP5 Ebyte 20dBm 2100m SPI NRF24L01+PA+LNA", not these red ones (without PA and without external antenna).
Of course we would like to have comparable results.

Yesterday I found out that if you think that you have a good power for NRF24L01+PA+LNA, just because "other model NRF24L01+PA+LNA workw fine" it is not the case.
Even

After not getting good results with external switching power, I replaced it with linear regulator based on AMS1117 and was pleasently surprised that it helped in my case.

Now at least cheap ebay NRF24L01+PA+LNA works. But for 5 same instances I built, I have different quality.
I will try to compare it with no antenna cheap ebay NRF24. Initial measurement shows funnny results where best seams to be gateway with antenna and nodes without external antenna.
Of course it is nice since the price, but what we all wish is to have as good as possible gateway / repeater and as cheap and smaller nodes.
So if red model without ext antenna or your shielded model is better at least for gateway... that shoud be nice.

I forgot to mention that I am interested of inter building communication (hotel) with as least gateways or repeaters as possible.

Very interesting. Can someone tell which ebay item is genuine red NRF24L01? Or other source with a reasonable price with delivery.
Also if possible PA+LNA version

I cannot see any more activity on this topic, but I have the same situation.
About year ago I bought ebay model with antenna and it worked perfectly and solved my range problem for gateway.
Now I ordered again from ebay (2 different stores, similar price) but on same arduino , when I replace old NFR24 with new ones (I got 5 different NRF modules) it doesn't work properly.
Also if I put NRF24 without antenna it works fine,
None of that ugly fix worked for me (shielding with alu, external power supply....)
Please note that old NRF24 work without any "ugly fixes", directly connected to Arduino Nano 3.3 (3uF capacitor only).
Also note that NRF24s without antenna work even without capacitor.

So my question is simple: can anyone differ good ebay NFR24 with antenna from a nonworking one. I spent not only money but also a lot of time (waiting items from ebay that dont work) without success.

Now I only can cry that I didnt order more pieces of that "old" model I had.
But I hope someone will be able to tell.

Currently I got 5 nonworking and 1 working with antenna, and 10 working NRF24 without antena.