If you're currently using an ASCII version of an MD5 digest, you could easily compress 1/3 of the space out by converting it to binary and then using BASE64 to convert to a string. (Or, even better, do it directly.) The ASCII version uses a byte to hold 4 bits of information, while BASE64 will store 6 bits of info in a byte. If you don't mind using a binary string, that would let you compress to 50% of the original size.

Other than that, I wouldn't expect much. Compression routines (at least the ones I'm familiar with) simply squash repetitive patterns out of the data. A digest tends to spread values evenly through the result space, so there shouldn't be much repetition for a compression algorithm to take advantage of.