GLBSE uses BCrypt + salt

I'm disappointed. According to LeakedIn my password is not part of the leak. It would have been interesting to see if anyone managed to crack my old password: h0NOl&tHgNr7ePTiayf7

Check This out.
http://shiflett.org/blog/2012/jun/leakedin
Link to Chris Shiflet's blog and another link to "Leakedin"
Their leaked password checker. 

Happy Hunting....

Only if it isn't in a dictionary somewhere already. But yes, even dictionary cracks are slowed down, somewhat.

I assume you mean Salted Bcrypt w/ workload=20, that is 2^20 = 1 million iterations.  Slightly harder.   A single round of bcrypt takes roughly 5x the clock cycles as long as SHA-256 (OpenCL optimized).  Thus bcrypt(20) is on the magnitude of 5 million times harder to crack than salted SHA-256 hash.

Another way to look at it.  If a hacker could brute force a given password hashed SHA-256 in 1 second it would take them 57 days on bcrypt(20).

There is absolutely no reason to use anything weaker than bcrypt (or similar chained iterative functions like PBKDF2 or scrypt).

pass - stupid
MD5(pass) - cryptographically weak
SHA-256(pass) - vulnerable to rainbow tables
SHA-256(pass.salt) - vulnerable to brute force
bcyrpt(pass,salt,2^10) - vulnerable to weak/common password list
bcyrpt(strongpass*,salt,2^10) - computationally infeasible to attack

strongpass being enforced by the site as
8+ char
not in dictionary
not in known password list

Would someone please explain this for the uninitiated: is there only one unique string (password) that corresponds to a given hash?  I believe the technical term is collision resistance, right?  Once you reverse the hash, can you know for sure that you got it right? If password is a dictionary word, it may be obvious, but how about if everyone were using random strings for their passwords? Would the hacker ever be able to know for sure if the reversed hash is the right one?

Yes, thank you. Now, is this statement still true when a typical password is shorter than the 32-byte hash? 

If you restrict the inputs to being within some normal distribution of user password length, then there are no longer an infinite number of inputs. So there are no longer an infinite number of inputs that can result in the same output.

That doesn't make the statement false though because of the hedging word theoretically.

One other thing I would note here is that the act of telling the public how many rounds there are in your password hashing settings may save the attacker quite a bit of work.

LastPass has your encrypted passwords. They don't, however, have the decryption key.