Since P5 refers us to RFC 3066 when in doubt about how to use the 
xml:lang attribute (as indeed does the latest version of the XML 
specification), and since RFC 3066 prescribes

    2. When a language has both an ISO 639-1 2-character code and an ISO
       639-2 3-character code, you MUST use the tag derived from the ISO
       639-1 2-character code.

the conclusion seems inescapable that when our documents contain 
multiple languages, some of which have a ISO 639-1 2-character code and 
some of which don't (such as ancient Greek), we will need to use both 
two- and three-letter language codes in a single document.

Not everyone feels comfortable with the thought of mixing these, and I 
too was initially surprised at finding the above in RFC 3066. How do 
others feel about this?