Proof of Stake is a proposed modification to the Bitcoin protocol.
maaku claimed that it doesn't require a hard fork / alt chain.
Is this true?
It depends on what PoS system is considered (or not, since I think the answer is the same either way).
Cunicula's system is completely different from what we have now. There's no way to "half" switch to it since a block which is valid in the current system means nothing in his system, and vice versa.
My system uses a skeleton which resembles the current system, so in theory you could solicit signatures in parallel and let anyone use them or not as he pleases. As theymos says a fork will still occur once there's a disagreement between the two systems about which is valid; and anyway, I think the process of collecting signatures and embedding them in the blockchain requires enough rethinking of the network that it is meaningless to regard it as anything but a hard fork.
Of course, even if you make an entirely new alt for it, you can always provide a built-in mechanism to allow bitcoin owners to import or copy their wealth, to make sure the early adopters of decentralized digital currencies at large don't suffer from the advent of a new and potentially better alternative to the original Bitcoin.
If proof-of-stake is adopted and someone succeeds in attacking proof-of-work but doesn't meet the new proof-of-stake rules, the network will split: old clients will go with the attacker and new clients won't. So I would say that proof-of-stake is a hard-fork change. A fork might be avoided if no one tries this attack, but you can't rely on that.
This doesn't mean that you can't do the change without causing chaos, though. A totally backward-incompatible change occurred in February without much turbulence, for example, though this change required two years of planning in advance. A similar gradual roll-out could be done for something like proof-of-stake.
(I don't actually think proof-of-stake is a good idea.)
Depending on the implementation, it might not require a hard-fork (which is defined as old clients not following the ongoing blockchain). If all POS-enabled blocks also have valid POWs, old clients will accept them too. This is similar to the BIP16 rollout last March.
I'm not sure proof-of-stake is a good idea, since it gives even more power to computer crackers who steal a lot of Bitcoins, and forces miners to keep their Bitcoins online for mining. It also could lock a lot of smaller miners out.