There's also a minimum wire size, based on physical durability requirements. A significant number of wires in the harness already carry no appreciable current and are merely signaling wires, but they can't be infinitely thin.

Getting most of the power wires down into the range of signal wires is a huge benefit, because it means they can use common terminals and smaller connectors, but that's where it stops.

For very low current wires, I've seen non-metalic (nylon?) strands woven in with the copper for mechanical strength.

Yes, typically the savings would be in the weight and physical size of the wiring harness (as well as possibly allowing for tighter bend radii). You'd design for a max amount of heat generated by the wiring harness, or possibly a max voltage drop if that's a constraint. You don't need to do heat dissipation calculations yourself, there are standards like SAE AS50881 that do the heavy lifting for you.

Edit: Smaller wire is also cheaper of course. That's probably a pretty significant upside when talking about a mass-produced vehicle.

Good point and you're not wrong. One mitigating factor is that resistance isn't constant: In copper it increases with temperature, and I^2R causes temperature to go up, which causes resistance to go up, and you get something of a positive feedback loop. Temperature rise is one of the main reasons why standard wire gauges are specified the way they are.

If you use 4x higher voltage, the temperature rise is much less significant (16x less significant to be precise) so it becomes more or less a non-issue and you can treat resistance as a constant.