I've done it. Algorithm::FastPermute.
It can comfortably print all the permutations of ten elements in less than a minute, on my iBook. So it's almost four times as fast as Algorithm::Permute.
Since it's based on Matt Day's code, I should get his permission before releasing it properly. I also wonder if it could be incorporated into List::Util. Time to write some email.
Update: Email to mday@icon.com and mday@lachman.com is bouncing. Does anyone have his current email address?