Well, if there's a GHC code performance bug then we should fix it, but I
don't think we should use a workaround in the libraries that gives a
function the wrong strictness.
The generated core has the strictness we want, the Haskell just doesn't reflect that (which means it could break in the future). I think we should stick with the current definition until we can come up with a better one that still gives the right core.