Not much successful day today. The first thing I realized when I woke up today was that the node from yesterday of course draws much more than I expected. I left the debug outputs to serial port in the code, followed by some 50ms of delay to make sure the debug outputs get transmitted before the node eventually goes to sleep :) Another power hog turned out to be the LDO. I sort of suspected that the datasheet does not lie when it specifies 17uA of quiescent current, and I was right.
I stopped working on it for the moment and ordered lead acid acumulator for further experiments. In the meantime, I just finished building a basic CR2032 powered node with bare 328p and using internal RC oscilator. Tests on modified Arduino Pro Mini were succesful, but the node doesn't work yet. I'll look into it tomorrow.
Update: I went shooting tanks, and quickly realized I slightly deviated from the standard Arduino schematic - forgetting the AVCC, and no pull up resistor on RESET. There should be an internal one so I thought WTF, and omitted it. Anyway, it turned out the resistor is important :) Now I have upload working and I can sleep better.