But there are other types of flash that are better suited for this environment. I see parallel and SPI NOR flash a lot in ruggedized embedded devices (usually with a minimum of 100,000 erase cycles), but they're generally a lower capacity.
I don't know the exact cause of this failure, but there are ways to design around unreliable flash too. They could partition it such that the firmware image is in a read-only partition, and it runs out of a ramdisk, with only user settings and log messages stored in a R/W partition. That may be what they're doing, and perhaps the issue is due to the software crashing when the write fails. I'm not sure.