2020 has been a good year for hitting framerate targets. Temporal reprojection (with dynamic internal res) & frame time estimation is doing an outstanding job over a range of AAA engines (including on new consoles - still worth paying for overhead; push hard then adapt with dynamic rather than aim lower).
Definitely need to have a go implementing it myself in 2021. Catch up on new timing feedback calls & maybe see if you can use VRS to lower frame cost rather than uniformly scale resolution?
When I first thought about the huge boost in asset streaming bandwidth in new consoles, I expected in-engine cinematics to cut more often & do far more cuts between locations. But I now think about how there is a subtle technical pressure (minor but it exists) the other way, to reduce use of camera cuts due to temporal sampling. God of War [2018] never had a frame where it couldn't rely on a decent buffer of previous frames that shared on-screen elements (improving consistency of TAA).