There's already an IETF draft that covers most of this: http://tools.ietf.org/html/draft-nottingham-http-problem-06
And it's much more mature, though it doesn't handle the "success" responses, which, in my experience, aren't necessary as the 2xx is usually sufficient.