The "IMPOSSIBLE TURN" document that HugMonster quotes is hopelessly pessimistic in its assumptions. In particular the claim that a tighter turn offers no advantage is something that the original author clearly didn't bother to support with the nitty-gritty aerodynamics.
The optimum angle-turned/height-loss comes with 45 degrees of bank at maximum lift coefficient. That suggests that a practical application of the manoeuvre would have one operating as close to stall as possible, perhaps on the onset of the stick shaker. By holding the AOA precisely, the height loss for a 180 degree turn should theoretically be just hundreds of feet even for WWW's 737, though the height lost depends on v^2 for the speed corresponding to that max Cl.
I have no reason to doubt the accident stats quoted though -- clearly it usually
is substantially safer to pick a reasonable landing spot straight ahead. I can think of a just few airports though (Antwerp, Tempelhof spring to mind) where trying a turnback from a few hundred feet in a light aircraft might just be preferable to a forced landing in the environs.
Dave Rogers article (cited by Code Blue) can be found
here.