For the moment:
>* your proposal requires all HTML documents to have a document
>character set of ISO 10646. I believe this is gross overspecification.
but I will emphasise: I do not require a parser to understand
all of ISO 10646, even for numeric character references. In
addition, my proposal is aimed solely at HTML. SGML in MIME
is much more sophisticated, and offers much better support
for the standard SGML mechanisms.
Perhaps overspecification is the price to pay for generality. The price
you extract is incompatibility.