Logged In: YES 
user_id=31435

If 250,050 balls are tossed into 2**32 bins at random, the 
mean number of occupied bins is ~250042.7 (== expected 
number of collisions is ~7.3), with std deviation ~2.7.  While 
we're happy to get "better than random" 
distributions, "astronomically worse than random" distributions 
usually predict more of the same kind of problem reported 
later.

That said, I certainly agree that would be a major, easy, and 
cheap improvement over the status quo.