Via YARGB (Yet Another Really Great Blog) [1] are some interesting password recovery times [2] based upon password content, length and type of computing resources one has.
I remember back in college (Computer Science Department at Florida Atlantic University) [3] (in the early 90s) we had access to a MasPar [4] with, I think, 4,096 processing nodes. There was talk of writing a password cracking program for the machine, which was a perfect use for the machine, being a SIMD (Single Instruction Multiple Data) architecture (same program on each processing node, but different data). The default Unix password scheme (at the time) used a 12 bit number to “randomize” the password, so there could be 4,096 different encryption results for any given password. A perfect fit for the MasPar—instead of having to do 4,096 serial encryptions of a guess, all 4,096 possible values could be tested at once. An incredible increase in speed (it could do in an hour what it would take a conventional computer about 24 days to do).
But alas, we never got around to it; I'm suspect it was because no one really wanted to program in FORTRAN (Formula Translation).
[1] http://yargb.blogspot.com/2006/04/tuesday-tech-digest_114472457798478649.html
[2] http://www.thecrypt.co.uk/lockdown/recovery_speeds.html