I've just put up a new draft, with the following changes:

  - I've fixed the examples to be clearer and to include all of the
    weird or peculiar cases.  I've also added some examples of invalid
    input.
  - I've included the formal specification, based on what Al* Petrofsky
    suggested.  (I'd like to note at this point my distaste for using
    such cumbersome tools as BNF-style grammars for describing Lisp
    syntax...  I attribute the complexity of the change to that, not to
    the complexity of this SRFI; the change to the recursive-descent
    parser shows how simple it really is.)
  - I've included Scheme48's reader, and I've fixed the slight bug in
    the #; definition which caused the semicolon to not be immediately
    consumed.

Regarding Aubrey Jaffer's use of #; for a different purpose: since the
use of #; for S-expression comments is already well-established, and
since it has already been decided upon for R6RS, I think it is better
for this SRFI to stick with #; and for some other character sequence to
be selected to substitute for SCM's somewhat obscure use of #;.

On the matter of nested comments: I am pretty firmly set on the current
way S-expression comments work.  Since it's an extremely specialized
case, since it has a very simple logical explanation, and since the
current way things work is a very simple addition to a traditional Lisp
recursive-descent S-expression parser, I'm inclined to stay with the
status quo unless the suggested alternative can be shown to be as
simple, in terms of logical explanation & recursive-descent parser, as
the status quo.