For fast "data-oriented" parsers of other formats which contain numbers (which to be fair are uncommon), number parsing is often the slowest part. For fast data transformations in the genre of jq (but not jq specifically), parsing is often the bulk of the runtime.
Not a downvoter, but something doesn't need to necessarily be a bottleneck to warrant speeding up. Put another way, why deliberately pick a slower method if a faster one exists?
Many applications these days suffer from "death by a thousand cuts" - there's no single thing which makes it slow, just lots and lots of slightly slow things piling on top of each other.
