Rendering complex shaded models to a display buffer MIGHT benefit from additional GPU power. Nothing requires more VRAM than you already have -- it is there in that large size to be sure you do not run out, and with 8GB already, there are vanishingly few workflows that could use it all up.
--------
Once an image is rendered into the display buffer, the amount of display RAM required to hold it is just the [embarrasingly small] amount of RAM for one display worth of data. (a 4K display takes on the order of 4K wide by 2K tall or about 8KK or around 8 MegaBytes of display RAM. Pitifully small. You have 8,000 times more than that on one card.
Your assumptions about how much 'power' it take to REFRESH a display (once its picture is complete in the display buffer) are WAY off base. Because the data rate is too fast for the processor to handle every Byte, this data is handled like high speed transfers to a Disk Drive. There is one Hardware Automation on the display card for EACH display that can be supported (so SIX in your case).
The processor must set the display buffer address, and the number of Bytes, and tell that Automation to GO. When the transfer of one screenful is complete, the processor get a notification, and sets it up again for the next re-draw, about 1/60-th second later. The GPU is not involved in any way.
The load on the CPU and GPU is completely NEGLIGIBLE -- so small you can't even measure it easily.
There is NO BENEFIT to having a second card installed "to drive more displays". You already have more than enough hardware to do the job you need, without slowing down in any perceptible (or even measurable) way.
You already have what you need. A second card will not be helpful in any way.
it is very complex to get a second card to work together with the first card, and you don't need it.
