Hi, I don't know if it's nice to send an "international" XML document to the user. If the document is an article, I, as a user, don't wan't to
download a i8 document for reading only one language. If you have a document that has French, Englighs, Spanish, German version then you have to downlad a document of 4n bytes to read just n bytes.
On the author side I think it is a good idea to have
i18 documents but I don't think that is bad to mozilla to hide the tag that is inside a tag with a lang different of the one you are seeing right know. That is because I can think of(In LaTeX)
"Let $a$ be a number, $f:\Reals\rightarrow\Reals$ a function"
"sea $a$ un numero, $f:\Reals\rightarrow\Reals$ una función" so if you have the first in one tag with the lang="us" and the other tag with the lang="es" then it's good that the rendering engine
hides the math. It 's a bit diffuse, but I hope you get the idea.

What about entity replacement and using an server content variation to hand the client the correct set of entities? (Of course, this doesn't work with a document saved-to-disk, but will people really care?)

Good points; I rather favor the point Warwato made. Namely, why would the server send a multilanguage document down in the first place?

Server-side processing for the HTTP Accept-Language header makes a lot of sense, though. Is there a standard PHP- or Perl- procedure for filtering a document by language? (Or do I get to write one when I feel like it? ;) )