> It seems like most recent performance gains come from runtime/dynamic optimizations, not static/compile time optimizations.

Are you sure? Have you heard of stream fusion for Haskell?