They're hardware accelerated so it's not worth making a new codec until you have a big improvement over the prior baseline, because it takes a long time to manufacture and roll out devices that are better. Verifying an optimization is worth it requires testing against a big library of videos using standardized perception metrics, it requires ensuring there's an efficient way to decode it in both hardware and software, including efficient encoding. It's easy to improve one kind of input but regress another. Most of the low hanging fruit is taken already. Just the usual stuff that makes advancing the frontier hard.