I think this thread got derailed in a major way. However, I could be wrong about what I am going to say, and that would be the case if I am wrong about this:
If someone were to try to crack a TrueCrypt volume that was either AES-Twofish-Serpent or Serpent-Twofish-AES by bruteforcing keys instead of a password, they would have to break three completely different keys that were all generated from the same password.
I believe I read that this was the case on Truecrypt.org or in Truecrypt documentation many ages and versions ago. The reason for this is so that a weakness in the outside encryption layer (flaw in AES or Serpent depending on which method) that allows an attacker to get the key won't allow the attacker to decrypt the other layers.
IOW, if someone managed to get the key (as opposed to the password) for the outer layer, decrypting the drive would return random data because they only decrypted one layer.
Now, assuming I am right about all of that, and depending on how it was implemented, I am not sure whether or not Horcon's original suggestion would actually be less secure than putting the two passwords together. An LM hash vulnerability would not apply because even if you get one password right, and even if it's the password for the outer layer, you're still only going to be returned random data, not confirmation that the password was right.
This assumes the passwords are strong enough on their own. If the passwords are each weak, then combining them to make a strong one would certainly still be better, but assuming the passwords are strong enough, I am curious as to whether or not entering a password for each layer would be less secure than combining all of the passwords into a longer one.