The regular expression matching used by Thompson in grep/ed may be slightly faster (it's really not "a million times" under 90% of datasets, that's just garbage) but it's far more simplistic.
I'm pretty sure Thompson didn't use his NFA algorithm in grep/ed.
So what exactly is your proposed solution here? The link you referenced doesn't provide one, it just says "PCRE is slow and it could be faster, theoretically".
has a functionally equivelant, and a considerably faster algorithm, when compared to
Perl's regular expressions. A solution doesn't necessarily involve entirely replacing things, either.
