is this a thread per “pillar” or a thread per region, you will also be making a big assumption on the number of cores/threads the user will have. Take a look at the Steam Hardware survey for core count, I can’t find the tread count, and we can not presume Hyper Threading/ SMT (because of different actions taken by Intel and AMD over the years)
maybe expanding the physical size and going into bigger “chunks” can “slow it down” it can also lead to more interesting deviations (especially if there is some random offset into the noise function for a given chunk).
be aware that there is a cost of spinning up the Async in the first place (segmenting memory which could hit paging with big operations or “many” small operations), and then a cost in the main-thread being paused on the completion of the Async. Then anything dealing with “spawn” happens on the main thread in the first place.