"This parser can be expressed in yacc" does say *something* (albeit probably not as much as people think it does) about the grammar itself - sort of a static-check on the possible ambiguity. Similarly, it reduces certain kinds of mistake (though it trades them for others at a higher level, mostly.)

On the other hand, the BSD ftpd used yacc for the protocol parser - which seems to have been a mistake of near-sendmail proportions. And the one major change I made to the GNU C++ grammar (nested types) couldn't be done in yacc, and involved C code between the tokenizer and the parser. I haven't had much excuse for writing parsers, in general - there's almost always a "simplify this to something I can throw at some pre-existing parser" path that makes things easier for the end-user too.