Bug 98800 shows lots of time spent computing overlap regions, but we never actually test against them in this case. Should figure out if we can avoid computing them in the first place.
Test Case - https://testdrive-archive.azurewebsites.net/performance/robohornetpro/ It is pretty fast now a days: Chrome Canary 111 - 3.4xx seconds Safari Technology Preview - 3.12 seconds Firefox Nightly 110 - 1.75w seconds Do we need to tweak it further?