the ghc docs suggest using Doubles instead of Floats because GHC specializes Double but not float...I want to use floats to save memory and to test this claim, i grepped through the libraries for the SPECIALIZE keyword and find that with the exception of a few complex routines (which I don't care about), floats are just as specialized as doubles...is this the case, or am I missing something?

I think you're right, there are just as many specialisations for Float as for Double. I've updated the docs. Cheers, Simon