I would recommend
strongly against
Parse::RecDescent for this. That was written before the /g modifier existed in Perl and so every time it matches a token it makes a copy of everything that comes after the token. On even a fairly small data set this can take a prohibitive amount of time and memory.
Changing that would entail rewriting the whole module. TheDamian had plans to do this, but I don't know if it ever happened. He did tell me that said rewrite was going to have to be incompatible with the original in some ways.