Apologies i was ooo, no need to be sorry to insist, this is great feedback.
A bug report with a link to the benchmark would be great. Also, make sure the abi change is actually invalidating kapt (the thing i mentioned above).
Also, wrt stubs, that task is incremental as well so it is never certain that KSP will always beat kapt. KSP 's impact is more clear for clean compilation.
Nevertheless, it would be great to see a bug report with a link to benchmark so we can reproduce.