The point seems to be really about avoiding the need for lookahead.
Parser structure simplifies a whole lot once you can throw away the
end marker, e.g. if the end token is just an end token, and not also
part of the beginning of the next symbol.