This is certainly a point that should be clarified in the next version of the
spec. My own interpretation of it has always been that multiple smaller
or bigger tokens should indeed have a differential effect, but it never
even occurred to me to think that multiple underline tokens would imply
double underlining. Since double underlining is probably not even possible
on lots of systems (e.g. video terminals), I'm inclined to say this is not
the case.
As for the two different ways to represent text that goes from large to
small to inbetween and back to large, they both look fine to me. I would
think that a good richtext parser should be able to display either one,
and that it is acceptable to generate either one as well. -- Nathaniel