The XML file has to be parsed at startup. If it’s large, it causes a delay at startup.
Then of course, many rules have their cost when checking text.
I remember measuring LT startup time in several language and several releases
a few years ago. I should probably do that again. Actually, using LT with my Vim
plugin, it seems that LT startup time is slow now, but I have not measured.
Comments are probably cheap to parse and not loaded in memory so I would not worry about those.
I’m pretty sure that XML files are loaded lazily, so if I only check grammar in French for example,
I won’t be penalized with large Dutch XML files.