Re: Accept-Charset support

Personally, I'm fond of the 'accept-hash' proposal for header
compression: the client creates a hash (e.g., MD5) of all of its
characteristics (accept, accept-language, accept-charset, user-agent)
and sends it in the request.

Non-negotiated resources just ignore it. Negotiated resources will
look up the hash, and if they've seen it before, apply it to the
content negotiation; if they've not seen it before or forget, they'll
ask for an expansion (and, hopefully, remember the configuration for a
while). It adds a round trip to the first time you connect to a
negotiated resource, but avoids the round trips thereafter.