Paralleling wires is stable because the TCR of copper is positive. When one connection carries too much current compared to its peers, it will heat up. This will increase its resistance, causing it to accordingly carry less of the current. So the system is self-balancing.
Do not remove ground wires. That is stupid. You'll just be raising the current in the remaining wires. EMI should not be a major concern as we are talking about DC power delivery here (also why I'm saying "resistance" instead of "impedance") and so the potential for trouble by changing the number of conductors making a connection is limited. Yes, anything could happen, but that's just the nature of EMC problems.
Yeah I realized that was the worst way to go about testing that anyway right after I went to bed last night. If (big stress on if) that was the issue a ferrite bead would be a better way to test it. Based on what you're saying my SWAGs were wildly off. I'd still like to see the sims of it however to see if they provide any illumination on the issue. What makes me think something weird is going on is that it's two out of six wires heating up to absurd degrees. Of the other four two are carrying normal currents and the last two (based on Roman's video) are carrying practically nothing. Buildzoid makes the convincing argument that clearly Nvidia engineers were aware of something like this could happen on the 3090. But, then didn't carry that over to the 4090/5090.
Do not remove ground wires. That is stupid. You'll just be raising the current in the remaining wires. EMI should not be a major concern as we are talking about DC power delivery here (also why I'm saying "resistance" instead of "impedance") and so the potential for trouble by changing the number of conductors making a connection is limited. Yes, anything could happen, but that's just the nature of EMC problems.