We now know the cause of the problems with the PIC16F648A. The internal EEPROM writing procedure damages some bits in the port status register, most probably bit RP0 is affected. And seemingly this happens only if multiple interrupts occur during writing the EEPROM. However, this doesn't really matter, the chips are not usable anyway. Writing the EEPROM takes about 4 ms, too long to disable all interrupts during that time.
regards