Yow... I was under the impression that all major regex engines used NFAs converted to DFAs lazily, with fallbacks to a slower engine for features that cannot (

Yow... I was under the impression that all major regex engines used NFAs converted to DFAs lazily, with fallbacks to a slower engine for features that cannot (or cannot practically) be implemented using an NFA (unbounded backtracking, that sort of thing.) What is the advantage to doing this?