But why? You'd only get partial corruption detection.

As I see it, either you have a lower layer you can trust, and then this would just be extra overhead, or you don't, in which case you'll also need error correction (not just detection!) for the database file itself.