Crypt::Checksum::CRC32 (among others) is implemented in XS, and unlike Perl's internal hash function, which may change from release to release and build to build*, is standardized. Also, when talking about speed, it's important to be specific and to benchmark. How fast is your current implementation? How fast do you need it to be? Have you profiled to find your bottlenecks? What kind of data are you checksumming (what size)? And so on.

* Update: Not only that, since hashing is randomized, it (clarification: in this case the checksum, not the function) will change from run to run, as I showed here.