Its not R/W cycles. The chips had a few cycles during initial development then after everything was working the circuit was put away. Bringing them out years later show several not working and needing reprogramming. Part of it might be that I live in the tropics and that room is not air conditioned every day.
One trick that I did not consider until now is over-programming the flash, something I used to do with EPROMs -program them more than once with the same code and no erasures in-between to make sure the floating gates were fully charged or not. However waiting several years between tests makes for a very long test. Don't know how many years I have left ;-)
It might be that the early AVR chips were not as robust as the current chips. I think that flash memory in general improved a lot in the first decade of this century as have the processes.