All things being equal, I think I'd be happy with the current encoding
method (name=value&name=value with escaping) just slapped into the
body of an HTTP/1.0 'SUBMIT' request -- keeps things simple and
straightforward, avoids creating a new syntax, is known to properly
handle escaping issues, and servers and code already exist for
decoding/handling it.
To be honest, the idea of encoding it in some kind of SGML format does
not excite me in the least -- what would be the added value?
Also, I'd like to suggest that multiline text fields have a different
type (e.g. "textarea") than single-line fields ("text"), as they
really are different types of things, and imply different issues
(semantics) for client-side handling.
Cheers,
Marc