This sounds like the kind of operations that a relational database is intended for.

I would suggest using DBI together with DBD::CSV if you want to keep on using flatfiles. And then when you get tired of bad performance you can just use a real database instead. :-)