spinning up threads and context switching, just like the cost for everyone else?

The Haskell version is provably going to work without issue when parrellelized (due to monoid laws), but I doubt the concurrency primitives in C is as trivially usable, and I expect no sane person would want to write that version.

There's nothing stopping you from following the monoid laws in C. There is also nothing stopping you from breaking the monoid laws in Haskell.

OpenMP IS trivially usable.

Why? This is not a hard problem and there's nothing wrong with parallelism in C. I'd much rather work on parallelizing the C code than deal with any amount of Haskell. I suspect there are more people who would agree with me than the total amount of people happy to write Haskell.

Whatever the cost in C, Haskell is incurring the same cost, unless it is using green threads.

(You could use green threads in C - or something similar, though to be honest counting words in a file is a bit of an annoying problem for parallelism)

Haskell does use green threads. The thing is, Haskell code is managed by a Haskell runtime, and each runtime manages a single (processor) thread. So the usual parallelism model for Haskell is to spawn a runtime for each processor core and then let each runtime manage its own green threads.