So it looks like this is a problem with cygwin only.

That makes some sense. When I briefly played with cygwin close to a decade ago I noted that it was very sluggish for many things. It wouldn't surprise me to hear that it has never been tailored for a multi-core, multi-threaded environment.

You might get better responses by reporting your findings on the cygwin lists as this doesn't seem to be a "perl problem" per se.