My current method is O( (n/h)^2 ) where h is the hash size, which is really O(n^2) for n>>h. I'll have to think about this to see if it would be better.