It's not advisable to use MD5 for anything. It's security is severely compromised. Weaknesses in MD5 have already been exploited to create a fake root CA[1] and a fake Microsoft code-signing certificate[2]. There isn't practical publicly-known preimage attack, but there was a theoretical attack published in 2009[3]. Additionally, MD5 can be calculated far too quickly to use for password hashing. A commodity GPU from 2009 can calculate 200 million MD5 hashes per second [4]. For password hashing you should use something like bcrypt or scrypt which is specifically designed to be computationally expensive.

[1] http://www.win.tue.nl/hashclash/rogue-ca/

[2] http://blogs.technet.com/b/srd/archive/2012/06/06/more-infor...

[3] http://link.springer.com/chapter/10.1007%2F978-3-642-01001-9...

[4] http://bvernoux.free.fr/md5/index.php

With many, many iterations, simply iterating a decent hash function is a valid 80% solution to the problem. 5 iterations of a cryptographic hash function won't buy you anything at all, though.

Hashing a hash is not generally considered secure because you have to assume that if your system was compromised, the hackers will know what methods you used, including the list of salts. If the hash runs quickly then you didn't really cause them any more time/work.

Assuming "several iterations" means "a million or more iterations" then you have captured most of what bcrypt gets you. You've broken their rainbow tables and they have to brute-force to find users using "passw0rd". You can tune the "several iterations" the same way you can tune bcrypt.

That said, don't roll your own. You probably screwed it up somewhere. Just use the bcrypt library call. Or scrypt to let you roll +2 against GPU attacks.

That's right! KDF seems to be better than hashing in the case the server gets compromised.