Raster is believe it or not, not quite the bottleneck. Raster speed definitely _matters_, but it's pretty fast even in software, and the bigger bottleneck is just overall complexity. Nanite is a big pipeline with a lot of different passes, which means lots of dispatches and memory accesses. Same with material shading/resolve after the visbuffer is rendered.
EDIT: The _other_ huge issue with Nanite is overdraw with thin/aggregate geo that 2pass occlusion culling fails to handle well. That's why trees and such perform poorly in Nanite (compared to how good Nanite is for solid opaque geo). There's exciting recent research in this area though! https://mangosister.github.io/scene_agn_site.
