My experience has been that those who don't like RD have a mostly theoretical/academic background and are likewise similarly unimpressed with "simple" or "unsophisticated" code in general.
For the author his case against RD is not for the implementation. At the end of the articles he calls for LR reference grammars (I imagine both for fuzzy testing and documentation) to ensure that your perfectly usable RD parser with nice error messages does not miss some edge cases.
