I was speculating pretty strongly that the RAMDACs would be the same across all the card versions, because switching RAMDACs would require more engineering work... but it looks like several of these versions are maybe using different speed-ratings of a similar RAMDAC?

Since the overdrawing is happening in screencaps, and not just on the monitor, can't we be pretty sure that the issue is in the memory initialization / mapping being done (since you swapped in the ROM from a card with more VRAM) and not in the DAC/monitor timings?

I suspect that DeclROM surgery to keep the executable code the same but change the timings table would be the direction to go. I seem to recall reading a book or article about DeclROM anatomy some years ago...