Hi!
I have heard that the culprit was Lee and that the problem was some kind of unspecified data import problem.
Well, TIMESTAMP is magic field which has too much magic in it, and the magic was less controllable before 4.1. The first timestamp field would auto-update itself on both inserts or updates (if not referenced directly). Second timestamp field would be set to current timestamp only on inserts. So if you, say, reset page_random, and have TIMESTAMP field, you'd be ending up with all fields loosing time information.
I've never really felt the need to change it, personally. It's not like changing it would enable any extra features or anything. You can convert between the representations easily enough, in either PHP or SQL.
Um, yeah, there would not be much difference of functionality (you can cast, anyway ;-). The only question is efficiency - 8 (or 4 in case of usable 4.1 TIMESTAMP) bytes instead of 14, and somewhat easier range scans.
It is not that it is the most significant issue at the moment, hence it stays there for a while.