According to my copy of Handbook of Applied Cryptography, Matyas-Meyer-Oseas, Davies-Meyer, Miyaguchi-Preneel, MASH-1 and MASH-2 all output an n-bit hash (where n is configurable).
I'm left with the impression, however, that very little is known about the security of those hashes when n is as small as 64 - which leads me to think that truncation of the more common (n >= 128) digest outputs may well be as good as anything (and a very simple solution as well).