I'm not too sure if base64 encoding has a place in a comparison like this -- being the only method here which is two-way. Interestingly, my own tests on my local machine suggest that SHA1 is slowest, closely followed by MD5, then in about one third of the time comes in base64_encode again closely followed by the winner (for me) CRC32. Clearly, a different pattern than is shown above!

I'd like to investigate further with more thorough benchmarking, just out of curiosity more than usefulness. :)