I tend to lean towards "every application gets a new media type".
In case it isn't obvious by now, so do I.
This is a fine maxim, but it begs the question: what constitutes
a separate, distinct application, vs a modification, profile,
extension, or varation of an old one?
In the case where you're allowed to have a document that mixes
traditional HTML, MathML, Vector Graphics ML, etc., are these
separate "applications" or are they one "application" ("renderable
XML document")?
But this then begs the more general question of whether there is to be an
attempt to "design" the XML usage space, and if there is, whether such an
attempt has any chance of succeeding.
If the answer to this is "no, we don't want to try and control the development,
direction, and use of XML" then your question is basically out of scope. People
will design whatever they want and register whatever they want. The resulting
mixtures and granularity will be whatever developers decide is appropriate. And
in such a world I see little value in having an XML top level type. (Perhaps no
real harm, but little value.)
However, if the answer to this is "yes", then the task at hand becomes one of
coming up with an appropriate set of rules that people think will actually do
the job and will be followed. And in such a world a top-level XML type might
have some value, if only as a place to attach the ruleset. (Delegation of XML
registration would also be something to consider.)
I basically don't have an opinion on which way this should go. My one
observation is that the IETF at least has typically opted to try and stay out
of areas like this in the past.
Ned