[tikhonj]

Most of the Haskell optimizations are aimed at making high-level code faster, not absolute speed. Something like loop fusion is trivial to do by hand, so it doesn't confer any advantage on hand-optimized Haskell programs. And when you're comparing to C, you're in the realm of performance where hand-optimizing code at the expense of simplicity, maintainability or even correctness is necessary. Otherwise you wouldn't have considered C.

So high-level optimizations like loop fusion do matter, but only for the 95% of the code which is not entirely performance critical. It's great for letting you write pretty (and therefore easier to write and more maintainable) code but not great for squeezing out all the possible performance in an inner loop.

Considering that many GHC optimizations are of this nature, it's really impressive that GHC is still good at speed in an absolute sense as well. So if you desperately need to wring out an extra bit of performance, you can still do it in Haskell instead of having to use C. Sure, the highly optimized Haskell will be relatively ugly, but it's still better than C and much easier to integrate with the rest of your codebase.