Returning to the original poster's "runtime drops then climbs again" from 2019 that never got an answer -- in case anyone reads this thread from the top and expects to find an answer to the OP 's question (rather than that of the second poster, which attracted a good discussion!), I did some quick research and think I can explain that behaviour...

Testing MCM1 task start-up with fraction_done_exact set shows the behaviour described in the first post; take that option out for MCM1 and that behaviour doesn't happen -- it seems that there is something in the way MCM1 initialization reports progress that causes the "drop and climb back" behaviour observed during the period where progress is shown as 0.5%... (Another oddity of MCM1 initialization!)

Given that progress reporting is fairly accurate for MCM1 without that option, I'd be inclined to not use it!

Cheers - Al.

P.S. I can show my work if necessary, but this message is plenty long enough to respond to what is effectively the necro part of a thread!