I'm mildly curious to see some benchmarks. I'm not interested in running any particular benchmark myself. I have but small doubt that someone could produce an example that runs faster with threads.pm than with forks.pm. But I'd be surprised if one can't produce examples where forks.pm is faster than threads.pm (not on Win32, of course).

Reading more of the forks documentation shows several other motivations for its creation beside performance (other than the reduced memory consumption which is mentioned as a goal) so I'm less convinced that the IPC for sharing has been as efficiently implemented as I expected. Though I'd still be very surprised if thread creation is not significantly faster with forks.pm than with threads.pm. And the module's documentation also supports my expectation that forks.pm is more stable than threads.pm (aka "less buggy").