An Xorg instance needs to be running on each GPU (which is what nvidia-xconfig -a or --enable-all-gpus does) to use nvidia-settings to change clocks and fan.
That's exactly the reason why I dropped ideas of making thin system. For AMD I did a USB-bootable one that required 1GB USB flash only, and only 500M of this was the compressed file system, the remaining space was allocated for copy-on-write overlay fs to save changes.
Ok, i see that you and NameTaker have tried to minimize maximum system size. Your objection seems to be logical and the result of your experience and tests too.