The distinction between a matching explanation and a confirmed cause is helpful here. Have you tried the same small test file on a preview deploy to compare its cutoff with the local result?
Thanks, Abhinav. No, I haven't, and that's the open gap in the post. The ~120 cutoff was on production, and the reproduction only ran locally in wrangler pages dev. So the post shows a mechanism that fits, not proof that production uses the same parser. A preview deploy of the same synthetic file would answer that, and if you or anyone else tries it, I'd be glad to hear the result.