Citation needed. Not saying it's not true, just haven't seen it in practice. Usually if you want efficiency, you have to drop down to intrinsics or assembly, no matter how much the compiler is helping you.

Look at the publications here http://benl.ouroborus.net/

Also look at the Disciplined Disciple Language (related to Haskell).

Thanks for the links!

It's not a question of whether it "can" or not sometimes. Say there's a performance issue in your code. In my experience, I have a much easier time figuring out where the bottleneck is just by inspection with imperative code than with functional code (especially when concurrency is involved). Sometimes, transparency is more important than raw output.