Hmm, a grammar will usually build a data structure that takes much more memory that the original text file, but it sounds a bit excessive to me that a 2-MB file would imply a 33 GB memory foot print, that's really a huge difference.

My second point is that, from a quick look, your file looks pretty regular, I am not sure that a full-fledged grammar is really required (understand me, I am really in favor of using real parsers as soon as the input gets a bit complicated in terms of positions of tokens and the like, but here, I am not totally convinced that it is not overkill or over-engineering). From a quick glance at your data, I would probably have a couple of regexes to get rid of comments and then parse it manually using the semi-colon as a separator.

But that's only a personal opinion, I did not take the time to analyze your grammars in any detail and I have no idea of what you are really trying to do with this data, so I may be completely wrong. This was just premature guts feeling after a very brief look at your problem, it could be that just working an hour or two on it would lead me to a very different conclusion.

Please also note that I know a very little bit about Regexp::Grammars, but next to nothing about Marpa.