I'm not familiar with Pegjs, but other PEG parsers I've seen tend to use the packrat algorithm, which is suboptimal for regular languages, because it memoizes parses to speed up backtracking, and regular languages do not need backtracking.
For example, if you were to write a recursive-descent parser for JSON and convert it to a packrat parser, you will often find the packrat parser is slower.
Now, extended regex's include backtracking, and that's where packrat parsers can soundly defeat recursive-descent parsers: super-linear time parses can become linear time. This makes packrat parsers a wonderful "default choice" but if constant factors are important and your language is regular, you will want to look beyond packrat parsers.