I would not bother pulling this out until the week before RC1 if the standard has not yet been declared final.

Otherwise, -1 on keeping it under another name.  The only hashes we bundle should be standard ones as those are the only ones people will want to use in the long run.  We'd be saddled with carrying along a non-standard likely not widely used algorithm implementation forever otherwise.

even if sha3 isn't declared before 3.4rc1, people building 3.4 against a sufficiently modern version of openssl that includes sha3 (as i'm sure some version will) will still have access to the algorithm.

otherwise i'm sure someone will package this as a module on pypi for older pythons regardless.