Personally, I wouldn't load it. I'd grab DBI and DBD::CSV and interrogate it as if it were a database. And then, if that were too slow for some reason, I'd actually put it in a database (using the same DBD::CSV for import) - perhaps a temporary table, and query the database.

That said, thus far, I've not had to move off DBD::CSV - it is fast enough for my purposes.