Are you sure about this though? When a VM is extremely overloaded the diagnostic tools may no longer be reporting accurately as they too are being strangled. Perhaps you could check journalctl -b in the VM (once things settle, or after a reboot) to see if there is better retrospective info there.
Look at the tips here about improving mpv performance on Qubes:
Also, you can bump up the vcpu value of the Qube so it gets more CPU priority. That could make a big difference in the scenario you describe.