> why should we invent Yet Another Table Format when we could use a > subset of an existing one that is well established?
This is the heart of the issue.
Tables are complex, and their markup will necessarily be so as well.
Unless the existing SGML model is demonstrably deficient, it would be
folly to generate another.