@laund fair enough. It reminds me of the issue we had in
#GHC where basically throwing cores at compilation worked up till you had like 16 cores, but below that the bottleneck was the module dependency graph building inside the compiler so parallel compilation speed-up was non-existent.