Gee ... I've got no idea what these AVR-Freaks have encountered. Those are ATMEL (yechhh!) parts. Do they use a positive-going RESET?

Once upon a time, the PIC folks had reset issues, too, but they've managed to fix that without a supervisor, though their manufacturer does produce one.

Early MOTOROLA/Freescale parts (HC11, etc.) had EEPROM corruption issues too, even back when EEPROM was just for task parameters, etc, rather than for code. They managed to get around those without a supervisor IC.

When you reset your 805x, you expect the internals to remain unchanged, except for those things that are supposed to be cleared, PC, DPTR, some PSW bits, etc. A lot of things are supposed to be left alone, thereby enabling you to figure out what happened, if there was an anomaly, for example. If reset changes some things you don't expect to be changed, in FLASH or in other memory or status, is that acceptable under any conditions? I'd say not!

It's high time, IMHO, that the question of whether it's a reset fault or something else be rooted out. Despite the fact that the supervisor produces a 90% reduction in ONE of the, perhaps, multiple problems, there's no guarantee that it's done even that until someone tests, and rigorously so, a set of multiple hundreds for a significantly large number of iterations as to make the one or two failures per thousand, statistically significant. Until someone does that, we're all shooting in the dark.

RE