I know for sure that the expansion board needs to have an EEPROM chip that the EC on the mobo reads to determine which expansion board type it is and configures the computer accordingly. Here’s the repo which generates the contents of that EEPROM. There are two types - dGPU and dual M.2 boards (repo refers to them as GPU and SSD). I don’t know what EC does differently for those other than enable bifurcation for the dual M.2 board.
You would probably end up reading EC source yourself or teaming up with @James3 to create EC firmware that supports this new card, as EC does all the configuration of the mainboard to work with the expansion bay module.
There are also other EC_GPIO pins on the board connector which control power sequencing and most likely need to be functional for the GPU to work correctly. E.g. 62-GPIO0_EC; 63-GPIO1_EC; ID0; etc…
I wonder if it may be easier to create a bigger enclosure with more z-height for the GPUs? Though I wonder how much deeper can the expansion bay PCB sit before the interposer stops reaching the contact pads on it…
PS: I’m writing my posts with the intention to orient you as much as I’m able to withing the available information. My knowledge, though, is not particularly deep in those topics, so while I can point you in some directions, I cannot really offer a lot of substance. Sorry for that