If you go this route, I would start with something as simple as possible. Like a two port board that just passes unmodified messages back and forth. This would also be a great way to log traffic on both ports and better understand exactly which controller is sending which commands and when. Then you can start experimenting with modifying commands. I would plan to pass all commands unmodified except under specific conditions, and focus first on identifying the operating conditions during which you want to modify the torque messages and cook that into a specific logical test in code. Once that's working, you can experiment with different ways of modifying the message to see what works best. This approach should limit the amount of algorithm development you have to do, since the stock controller code will still be governing the vast majority of operating conditions.
Just remember, when you break the bus any failure in your software or hardware can do a lot more than break axles. It can result in a loss of your ability to control the electric drive. The result of which can be uncontrolled acceleration and death. You need to be very methodical, and have a specific plan in place to deal with that eventuality. Using custom code and consumer grade hardware in an automotive application its less of a matter of "if" failures will happen as it is a matter of "when." IMHO that's the problem with breaking the bus vs. spoofing. Your HW/SW has to work 100% of the time, there is no way to disable the mod once the bus is cut.
I really hope you can pull this off! I'm fantasizing about building something like a LoCost 7 or Morgan 3-wheeler, putting a Leaf drive train in a 1200-1400 lb performance platform where this type of mod would have a huge benefit