It's actually two bytes that differ. The last byte on each line is the checksum, which is part of the intel hex file format, not part of the data. The other two bytes form a 12-bit instruction word (note the '0' nibble in all cases; this is the unused upper four bits of the second byte).
As with Iam.M, I am concerned about the third case ending in FF0FEF. Ignoring the EF (checksum), the instruction word is then 0FFF, which is the erased/empty state of the EPROM. It is also the instruction "XORLW 0xFF" which will simply flip all bits of the W reg and continue. This should be a "MOVLW N" where N is a 8-bit number to be optionally used to program the oscal register. This instruction looks like 0x0C?? which is what the first two examples look like. If you do not use the internal oscillator then there is nothing to worry about, as you haven't written the code to take advantage of this value. If you use it, then you have a questionable device that has not had the value factory-programmed as expected. Is it perhaps a 12F508 device that was erased?