Hi, everyone.
I do have some experience with TEI markup (but don't know anything about the MW extension).
TEI is an XML dialect. As such, there are general XML tools useful for editing it, and one of the favorite tools among TEI practitioners is the (proprietary and commercial) Oxygen XML Editor[0], which has significant customized support for TEI, but is a desktop application and is not collaborative.
Lars asked a good question, in that the goals of those scholars are rather different from ours: TEI practitioners are general in the business of creating critical editions[1] and multitexts[2]. They use TEI to denote textual details such as emendations, corrections, crossed-out words, divergences between different textual "witnesses", scribal notes, sometimes even a change of ink or scribal hand; see for example [3]. They spend years and six/seven-figure sums creating things like [2] or [4] the unfortunately-firewalled-but-huge[5].
Whereas the TEI folks are generally:
a. academics and paid professionals
b. well-funded
c. radically detail-oriented
d. working until it is done/perfect
e. focused on depth of critical edition over quantity of works produced (dozens of man-years per critical edition)
We Wikisourcerors and gonzo librarians[7] are generally:
a. volunteers
b. not funded. (though we do have access to grants if we need them)
c. fairly detail-oriented
d. release/publish early and often; fix things as we go; tolerate errors and trusting an asymptotic improvement curve.
e. focused on quantity of works made accessible. (dozens/hundreds of _hours_ per work)
This, to me, suggests that there should be a reasonable limit to how far we go out of our way to support the TEI community on the Mediawiki platform. While a powerful collaborative Mediawiki-powered TEI editor would be neat, I don't think it's the Wikimedia movement that should be the main force driving (and funding) its development.
Hope this helps,
Asaf
[6] disclosure: I have collaborated and am friends with some of the engineers working on these two projects.
[7] To those who may not know, Lars runs Project Runeberg; I run Project Ben-Yehuda. Both are free volunteer-run digital libraries of public domain texts.