I have seen NVM corruption except for two reasnons: Bad code or bad reset, never the actual NVM. If you do not have a GOOD supervisor chip (with NVM I Insist on 4.75V cutoff and a chip specified to 4.5V) you are ashing for it. Even with this you can get an occasional erasure of a serial EE. There are ways to ensure good data:
1) 2 supervisors one at 4.75 connected to an interrupt, when taht interrupt happens hands off the NVM (except finish current operation), and an actual reset supervisor at 4.5V. 2) write everything 3 times, on read accept the 2 identicals, you namy get old data (one update lost) but it will always be valid.

Happy hollidays,

Erik