Your second approach is pretty much how SQL::Statement handles joins, it loads each table only once and then searches in-memory hashes.