FilteredRE2 (and regex-filtered) show amazing results on ua-parser work load, however if for some reason they're not available (or they're missed) users are stuck with the horror show that is sequential regex matching.
regex filtering should be reimplemented for the stdlib's re module in order to see how it behaves and if the gain is as significant as it is for the native regexes. The big question will be how efficient a pure python aho-corasick is (possibly via ahocorapy?)