Stuck buckets

[kosso_olli]

The headlights are just an example. We have the same issue in a lot of other situations as well.

[^Lele^]

Yep.
Headlights are stereotypical. They are slow by design. They are the impossible "chrome spheres" of raytracing.
The potential issues are very many, and there is no engine out there which doesn't suffer from some of them (unless, ofc, they entirely forego adaptive sampling, eh.).

I'm too old school, perhaps, but even with the "magic" coming, i'd say a user's best bet is to spend half that stuck-bucket time in scene optimisation.
They used to crop up when i was in production, and in a movie production they used to crop up a LOT.
At some point, for all the pipelining and formalisation of the process, a human (often me.) had to stay late to *manually* ensure the sequence rendered in due time, and to justify to supervisors the -hopefully minor- change in looks.

Notice that not all issues can be cheated.
Take fast moving, tiny highlights (CFR. fast spinning wheel rims.), for example.
There may be tech researched which could help, but they are inherently difficult for the statistical rendering we (all!) do.
Very unlikely to be found, very intense when they are.
One either suppresses them (and goodbye highlight) or one bites the bullet and samples them properly (ooooh, boy. that can be very very very long.).
If a human knows the shot will have those, then it can be planned for better (passes, pre-rendering compared to the rest of the scene, and so on), which may mitigate its impact on the process as a whole.

I'm being rain on the parade, i get it, but i still would not rely on the fix: it may be it'll only go so far, and that the results won't be good enough for everyone and all cases.
A stronger pipeline, and the artisan's love to one's task, will take one farther, and quicker, i'd wager.