> Content snippets should therefore be served as text/plain, the fact that
> the snippet *may* contain markup is irrelevant. The application is the

Opaque content should be served as application/octet-stream.  text/plain
should be used for material that doesn't contain markup.  That's what plain
means.

However, the arguments for the use of the external entity type make sense
to me.  I agree that the ones originally suggested are wrong.

> only tool that should have access to the content snippets, it's
> configuration should ensure how the snippet is handled (i.e. included
> verbatim without parsing it or doing anything else to it).

An application transcluding fragments might well treat plain text as 
CDATA, or as completely devoid of XML codes, whereas when transcluding 
XHTML fragments would parse the content.