Make the update step run in parallell

[oyvindeide]

We can look into ways of parallelizing the update step, options include looking into dask, openmpi, etc. Must make sure this does not make #6334 more complicated.

[oyvindeide]

Blocked by: #6474, we can revisit once we evaluate the results from that.

[dafeda]

Did experiments with using multiprocessing and shared memory, but could not get consistently better performance.
Note that I did not spend much time on this, and it is probably possible to get some gains here.
Closing this for now, but may re-open once we get some user-feedback.