First spec sheet I looked up from STMicro stated 1 million write cycles and 40 year retention. Second one from Microchip says a million as well. Third one from Atmel says a million cycles + 100 years retention.

But it's worth thinking about for a while to come up with a flash solution since even adding a $.10 IC adds up over millions of drives.

It's Mean Time Between Failure. Your example gives it as a hard number that none of the drives will survive past. There will be a statistical spread, but not like you describe it.

The actual spread depends on the shape of the failure distribution which isn't normally given. In a lot of cases, it's a Weibull curve which front-loads the failures and has a long tail. Even then though, you don't expect more than 60 to 70% of drives to have failed by the MTBF point.

You could also put a small EEPROM on the controller to store that information. A 128-bit EEPROM would probably be less than a dime in large quantity. A 1k-bit would cost just a penny or two more.