I run Milkyway, where the tasks take one minute on a graphics card. I can't download them that often, the server doesn't allow it. I only set my cache to 3+3 hours though. For example, on my CPUs I currently have 66 CPU tasks running and 12 tasks queued. It doesn't seem right to me to be constantly downloading 1 task at a time, I'm sure the server prefers I download batches of say 6. In fact on my 24 core machines, they could be trying to get tasks from Rosetta extremely often if they got one at a time. What if one task finished at 3:00pm, so it downloaded a new task, then another core finished at 3:01pm? And another at 3:02pm? I suspect I'd hit a server limit of how often I can contact.

I have no idea as I've never used it, but would it help to set something in Boinc's "Daily Schedules" tab under Computing Preferences? I wonder what it would do.

I don't know. It doesn't seem to forbid specific days. Just specific hours some days. In the meantime, i will try to reduce further the cache to see if it helps well enough. :)