It isn't clear to me whether your opposition is to the creation of new
top-level media types period, or just to xml/.
I'm opposed to creating more top-level media types than are necessary,
or to adding more parameters and other doodads than are necessary,
where I define 'necessary' as 'having a compelling application TODAY'.
So far, I've seen some good arguments for text/xml, application/xml,
a few calendaring applications and (possibly, if the issues about
subsets and compositions are elaborated), text/xhtml.
But I don't see any reason for creating new media types otherwise,
until they're needed. For the most part, applications can use
text/xml, application/xml, or even application/octet-stream, since
the MIME type is irrelevant.
I don't believe that "xml" fits into the guidelines for new
top level types, which seem to rest, not on the argument about
"default processing" but rather "device/gateway filtering".
At least notionally, the distinction was between text/ image/
audio/ video/ and (now) model/, with "application/" being the
catch-all for everything else. I think "xml" is, for the
most part, "application/".