Actually, it's a great idea, but it would take a significant amount of solid data to make it worthwhile. In the aerospace biz we employ thousands of people who take real service life data on almost every high-value part, and using things like Weibull analysis, they prepare reliabilty analyses which predict when the parts need to be swapped out. In the aggregate, such an analysis would consider the 100K blown-up engine as well as the more typical 250-400K lifetimes, it would be 'blind' to whether the car was in taxi service or driven by Grandma on Sundays. Failure modes are typically categorized as 'infant mortality', 'wear-out', or 'accident/random failure'.

But for us mere mortals, we tune into Brickboard and get empirical as well as anecdotal information so we know what to prepare for!