I agree with everything you just said.

Lets do XML and see how it flies.

The only Comment I'd make, is do we wait for the structure of the database file on the empeg to be documented (and read) or do we simply go with whats in the *1 files as a first cut.

I think *1 files is simple, bvut has a real performance penalty in that Mark Lord has to open and read every *1 file to produce the output.
the database idea has merit, but we need to have some working code that can work with this file first so that Mark can use it instead of (or as well as) the *1 files.

I will be interested in the proposed XML RFC when you release it [either draft or final].