No, I didn't seriously consider it. I have a minimal programming circuit with ZIF-sockets to program DIPs, but those are easily removed from a circuit (assuming they are put into sockets and not soldered) in case they need to be reproprammed . Doing this with a TQFP seems like a one-way road, so I think a way for in-circuit programming should be provided regardless. I mean, I didn't have to reprogram my first revision nodes so far, but I might want to add signing support later on or upgrade to MySensors 2.4 or 3.0 for some reason, so it should definitely be possible, but I'm not too worried, if the programming process is a little cumbersome, if the SOICbite experiment doesn't work out.

hello! can you advice how flash this fw in arduino ide with avr isp? not work with atmega without crystal yet

Forgot about this guy... Updated the project a little, the PCB now have footprints for both normal and PA+LNA modules. The old boards are still available on the github repository if needed. Even with the extra length added to the PCB with the dual footprint, it is overall still smaller than a regular non SMD NRF24L01 module.

@EddyG PCB received and tested, new version (v1.1) available here and on github.

Okay, finally got the original nrf24l01 SMD modules with the square chip visible... As exptected, things work just fine now, so the culprit of my issues was really the incompatible/fake "nrf24l01" SMD modules with the epoxy blob. If you want to use normal nrf24l01+ modules together with this smd gateway, DO NOT USE the smd module with the epoxy blob, as they won't work. Using only the fake SMD modules might work, too, but mixing the fake modules with real ones does not work.

I have been using an old(cheap) wall-to-USB power supply. I'll give it a try with a beefier supply and let you know the results I get.