Testing on one of our web servers showed current crypt takes about 8us while MD5 crypt takes about 1ms, which is over 100x as much CPU (and we'd need better enforcement on the cap on password length but with a limit much higher than 7; ). But my proposal was to not even include the full non-salt encrypted password in the cookie, just the last $n characters, say
20 . Then we aren't even calling crypt on every page hit (just when somebody changes their password). And somebody sniffing your cookie doesn't mean they can dictionary attack and find your password.
Corion and I have each worked on new "password reset e-mail" implementations.