Did you use a profiler like Devel::NYTProf to identify the bottleneck?

If not then it is worth doing as the standard output gives line by line timings which will pinpoint where most of the work is being done within each sub (how many times each line is called, how much time is spent on each line, etc.).

I've used Devel::NYTProf to analyse a Gtk2 GUI to good effect.