This is just a quick note with references to several TEI-L e-mails that may be handy wrt xml:lang issue (apologies for going first):
Piotr Bański on a general policy proposal ("accept and ignore the silliness that is xml:lang") and a script-related problem that doesn't vanish even if we agree that attribute values are symbols rather than words (note Paul Schaffner's example quoted at the bottom): http://listserv.brown.edu/archives/cgi-bin/wa?A2=ind1104&L=TEI-L&D=0&T=0&m=14690&P=10552