In rB6c302d6529ec: Sculpt: use TBB instead of BLI_task for multithreading we switched to TBB for sculpting, to avoid threading overhead issues on some platforms.
Rather than trying to make our own optimize task scheduler implementation, we should switch to TBB. If we use it everywhere, we can have a single thread pool for Cycles, Mantaflow, OpenVDB and the rest of Blender. It could also help improve performance in some areas.
An experimental patch from 2017 by Sergey is here: