Greetings,
I second what GrandFather said about devel profile. You might want to have a look at FFI::Platypus It should allow you to rewrite the routine pointed at by the profiler in a more optimised way (namely as a C library or even in assembler I guess)
Keep in mind that first I'd try and find more patterns in the data (as proposed by other posters) so I can get as much leverage as I can before doing a rewrite at a lower level