Benchmarking is often (always?) the best way to tell which of two implementations is the fastest. Something like:

sub create_parallel 
{
  ...
  # code here to create files using fork
}

sub create_loop
{
  ...
  # code here to create files in a simple for loop
} 
cmpthese(-5,
         parallel => \&create_parallel,
         loop     => \&create_loop,
        );
[download]

The -5 means that the subs will be run several times until 5 seconds of execution (each) are reached, so it might be a good idea to include cleanup in the functions.

I doubt you will find a significant difference though, or the parallel version might be the slowest. Unless you are writing 13 files on 13 different drives, or doing a lot of networking operations in each thread, each thread will basically just wait for the previous one to stop using whatever device you are writing on.