Thanks. I hadn't noticed the package in CPAN because it's not common to find the package you want so far down the list. However in this case I don't need to limit the number of processes, so I will forgo both Parallel::ForkManager and Forks::Super

So in the end, I am opting for plain standard fork as in the 1st example from kennethk, but, as suggested by mbethke with the inner loop and fork moved to the child of the outer fork (which is what happens in Eliya's example, I think)