Nothing supports 3.1 or 4.0. I will almost certainly never implement 3.1 (it was meant as a kind of "stop-gap" before moving to 4, but that's unnecessary). I've long considered implementing v4, but the current draft needs a few modifications, and while it adds some nice features, there really hasn't been a strong demand for them (versus video-encryption, which does not require a new format, and which Robert Vassen has started investigating).
If you're interested in pushing forward on a v4 implementation, I'd be happy to fix up the spec with the feedback I've received so it's in a place to implement. The changes are pretty minor. My great fear in actually moving forward on it is that it would lead to v4 encryptions that only can be decrypted by about half the implementations. So I'd be interested to see how many others are interested in pursuing v4.
The two big features are:
* You can quickly determine if the password is correct without decrypting the entire file
* You can tune the number of PBKDF2 iterations, which makes it possible to interoperate with JavaScript (which is too slow otherwise).
-Rob