If it's a contest of Marpa vs. regular expressions, I'm only surprised that it is just 3x. If your grammar parses using regular expressions, regular expressions will always be much faster.
I assumed when I saw your grammar that it was part of a larger one, one that was not convenient to parse as a regular expression. An advantage of having done it with Marpa is that you can extend the grammar much more easily -- but if that's not going to be the case in your application, yes, Marpa may well not be the best choice for you.