performance in base implementation is not sufficient to support this but we had some ideas of areas to start solving this problem.
Check out the cell where we configure a dask `LocalCluster here, it's just two lines of code
https://scanpy.readthedocs.io/en/stable/tutorials/experimental/dask.html
I genuinely don't know what dask's defaults are without one. But you can tune the number of workers etc.
https://docs.dask.org/en/stable/deploying-python.html#reference