Ghost currents with pwrd risers and 2 PSUs

Hello,

If you have good EE knowledge, I will consider a small donation to you if you bring me high quality answers to my questions. I have been googling and searching this forum for a while, I have found several pieces of answers but also it's often just best guesses. So no offense to those like me who aren't clueless about electricity, but I really need someone that can convince me that he knows what he's talking about.

1. Using two PSUs on the same rig (MB)

For now, let's forget about risers and suppose I'm plugging the GPUs directly into the slots on the MB.

My understanding is that you have to consider the 2 PSUs as separate +12V rails. I already have all necessary wiring to use PSU 1 as main PSU i.e. driven by and powering the mainboard, and PSU 2 is also switched on by the MB at the same time using this.

Supposing my GPUs have 2x 6/8 pin connectors, I'd better use the same rail for a single GPU. It seems natural, makes sense.

Now if I have GPU 1 and GPU 2, my understanding is that it's ok to connect PSU 1 to 2x 6/8 pin of GPU 1, and PSU 2 to 2x 6/8 pin of GPU 2. In this case, obviously GPU1 will be powered by PSU 1.

GPU2 is more interesting: If my mainboard is powered by PSU 1, then GPU 2 will be powered by both PSUs: up to 75W from the MB i.e. PSU 1, and up to 75+150=225W from PSU2 through the 6/8 pin connectors.

Is everything that I wrote above correct?

2a. The Riser Mystery: The Truth About Powered Risers

After reading a lot of various, sometimes contradictory sources, I would understand the following and/or have the following questions:

Powered risers are sometimes required, sometimes not, depending on total (all GPUs) power draw from MB and/or MB/GPU design
When powered risers are required, is it only to offload the MB, or can it also be because the power lines inside the ribbon cable are too thin to carry required amperage and would overheat / melt?
The latest powered risers that I have received only have 1 +12V line connected to the Molex. My previous batch had both +12V and GND connected, using a decoupling capacitor on the GPU side. I was disappointed, because I think it will only offload the +12V line and not the ground - if the reason for powered riser is ribbon overheating, the current will be sunk through the ribbon right? If that's not the reason, then it's still current sinking back to the MB instead of the Molex? Or not?
So the ones with the GND and the decoupling capacitor are better, am I right or not? (seems obvious)

2b. The Riser Mystery: Cutting The Cord

I have noticed that on the "new" powered risers (those with only +12V connected), pins B1 and B2 are only connected to the Molex +12V (not connected to the MB), then there's a jumper wire to pins A2 + A3 on the other side, but those ARE connected to the MB through the ribbon cable. I tested it with continuity check and it's positive. My conclusion is that both +12V from the Molex and MB must be the same line/voltage to avoid ghost currents between them
Am I totally wrong about "ghosts currents" and are they actually a non-issue? I mean if +12V Rail A (Mainboard) is very loaded and actually say +11.6V, while +12V Rail B (Molex/Riser) is at low load and actually closer to +13V (these are figures I got from my own measurements), then there is 1.6V difference between both ends of the riser and a very low resistance, that makes a very high current and it will eventually smoke, won't it?
Because I didn't want to take chances, I decided to cut the cord. So I cut A2 + A3 on the ribbon, so that the GPU would be totally isolated from the MB +12V lines. But I still have one question pending, it's about the ground. Since the powered riser only connects the +12V to the Molex, the current sourced by PSU 2 (Molex) is actually sunk to PSU 1 via the MB i.e. pins A4 + B4 + B7 of PCI-e...!?
The best way to avoid this would be a riser with +12V *and* GND, totally isolated (cut) from the MB, right?

Phew that was a long post... thanks to anyone that can help me, electricity is still a strange beast to me ;-)

PS: my 2-PSU rig is hashing fine, so this is mainly about optimization and peace of mind...