It is highly doubtful checkins are causing issues as they are not spamming the server very often (one check every 5 seconds per machine.) If it were a server spamming issue it would most likely come from progress reporting as that checks in every 3 seconds. However the slowdown on images is the main issue as quazz and Wayne pointed out.

While I suppose removing intermediary requests could help lessen the work stop/start, this would be primarily micro-optimizing as all the checkin does is update state, if needed, and decide if it is safe to start doing the tasking.

The conclusion that I came to on my own is - unicasting is heavy, and heavier if you run many many at once. It’s disk IOPS endlessly basically. Lower your maximum clients, use Multicast, get a better server with faster disks.

If they all (or several large groups) require the same image deployed, you might have a better time with Multicast.