https://github.com/gavinandresen/bitcoin-git/tree/fuzzer
... but it is pretty narrowly targeted at stress-testing the transaction validity code.
Last year at the NYC conference I mentioned that I would really like to see somebody who knows a lot more about network simulation than I do develop tools to help test proposed protocol changes, simulate Sybil or DoS attacks, etc.
I'd still really like to see that...
afiak Yifu from BitSyncom is to plan to test-execute a Finny Attack with Sybil elements on the prod-network after they get back from GDC, and documenting how hard is actually to pull off, and how it is generally a no-issue to accept 0 confirmation tx and talk about how to minimize risk.