You'll often see wild swings in UZR with no discernable difference in fielding performance from year to year, that's one thing. Teams with strong defesive positioning an artificially inflate UZR for players, and vise versa. The statistic takes an extremely long time to normalize and often by the time you get an accurate representation, the player has often lost a step or changed positions.
Those are a few of the many problems to take into account. It's not useless, but it is flawed and stated far, far too often as something that is always reliable and accurate, when it certainly isn't.
I can't wait for field f/x to be incorporated, then we'll get something truly useful.