Any solution in all of that?

Not one you'll like. I reached the conclusion a long time ago that using fork on Win32 is a pointless exercise.

The emulation simply isn't accurate enough, nor complete enough to enable transparent cross-platform solutions. And if you need to put conditional code in to test and account for Win32 limitations and variations, then it's better to move to a full Win32 native solution.

See this Re: Question: Fast way to validate 600K websites from a few days ago that does almost exactly what you are trying to do using Parallel::ForkManager. It works (on Win32 and anywhere where threads works), and the code is hardly more complex.