I can suggesttwo possible problems: Either the CPU has managed to execute the ISP/IAP routines, or the flash itself is bad.

Since the C66x parts do not fail, I would suspect the flash. Is there a common date code in the failed chips? Will the same part fail again in 1-8 days? If you heat the part, will it fail faster? Is it more likely to fail at a higher supply voltage? If the answer to these questions is yes, they you likely have some bad chips.

A typical manufacturing problem with flash is a leakage current that will cause a bit to discharge over time. Small imperfections in the gate oxide (too much silicon, not enough oxygen) allow charge leak off through tunneling. This weakens the oxide. which makes the tunneling worse. Eventually a short develops. Many manufacturers include a high voltage test (7-9 volts!) to force failures in these weak bits.

In the case where it is the unintended execution of the ISP / IAP routines, you can put a code trap (jump to the same location) just below the IAP routines (0FBFFh). It is better to hang up there than to inadvertently write to flash.