[Update for future generation]
I think I found my issue. It was the large transceiver see image below, with power amplifier and external antenna.
Once I replaced both receiver and transmitter with identical units, range got significantly better.
60m clear sight and almost whole house through multiple walls.

For whatever reason small unit can't hear ACK from the PA unit. I tried to reduce power on the PA end to

 radio.setPALevel(RF24_PA_MIN); 

but it did not seem to have any effect. My best guess now is that oscillator on the PA unit is slightly off. PCB antenna unit is battery powered.
Power source to arduino UNO that drives PA unit is good, 5V less that 2mV ripple. The transceiver is powered via 3V3 output (onboard regulator) of UNO though, could it be the problem?

Some ppl suggest - DIY.
The problem with DIY is that you need to know well what are you doing and you also need tools to debug your power supply when it does not work. Non of that applies to a typical novice.
I suggest getting one of the low-end bench grade power supplies ~$100 on amazon.
Example: http://www.amazon.com/gp/product/B00FPU6G4E

Make sure you don't try to save $20 and buy the cheapest possible with low precision pots.
Hi precision pots a re like $10+ a piece alone so you are likely to save nothing if decided to upgrade yourself. I did that and I regret it.

One bench power supply is not enough so you can DIY your second one.

One might think that to measure ripple noise of $2 Chinese power supply ~$70 unit is an overkill

What are the benefits of software controlled PSU versus turn-knob style?

@Oitzu DSO138 scope is like $25 from aliexpress

Some ppl suggest - DIY.
The problem with DIY is that you need to know well what are you doing and you also need tools to debug your power supply when it does not work. Non of that applies to a typical novice.
I suggest getting one of the low-end bench grade power supplies ~$100 on amazon.
Example: http://www.amazon.com/gp/product/B00FPU6G4E

Make sure you don't try to save $20 and buy the cheapest possible with low precision pots.
Hi precision pots a re like $10+ a piece alone so you are likely to save nothing if decided to upgrade yourself. I did that and I regret it.

One bench power supply is not enough so you can DIY your second one.

@Oitzu
I though antenna ground and MCU ground is the same thing

Typically you point it up. It gives you vertical polarization.
You point both transmitter and received wire antennas same directions.
It's possible that your chips are at fault, if non of other tricks worked (capacitors, clear power, enough current, short wires, lo-noise rf spectrum channel) try other transceivers.

It must be something very dumb. Check that you have a solid galvanic connection to nrf modules on all lines.

I think @carlierd means that resistor '102' (1kOmh) is serially connected with led. It's also on the side so removing it much easier - you just heat one side and push then another side and push away. In the reference design I think this 102 is actually R11.
Correct me if I'm wrong.

@hek Speech recognition is a fantastic idea, I've added it to my list of may be projects!

Is there any use for all this CPU power in home automation scope?
I struggle to find any use for them.