Disk I/O isn't relevant here since you're testing the performance of the parser, not the speed of reading the string from disk. I would expect a file this small

Disk I/O isn't relevant here since you're testing the performance of the parser, not the speed of reading the string from disk. I would expect a file this small to be read into a buffer up-front & the benchmark code just invokes the parse function on it. Indeed, the document is small enough, that I'd be worried about it hanging out in L3 cache the entire time which would overestimate the parsing performance which would typically see strings in random spots in memory being parsed.