No, no, you got me wrong. The point is in faster invalidation, not validation. That means you are immediately able to tell an invalid token from a not-yet-determined one (but not from a valid one). Then you have to hit the database, but the number of hits is - in some cases drastically - reduced.