Under best conditions back then, the highest I ever heard was @70% out of the second card, worst case scenario was in the negatives % since there was no cooperation between the cards, you'd pretty much get the primary card putting up its frame, and the second card put up nothing, ending up with 50% of possible fps. As seen in ACO. CF was worse.
I believe you can set individual profiles in NVCP per game, and thats what I'd do, set ACO for single card with a physX secondary, and in games that do take advantage of SLI, profile the SLI. It's not the best case scenario, but it's going to be the best the actual game engines will allow.
And no, it's not a simple thing just from the standpoint that vram is different, gddr5, hbm, gddr3, etc and they'll all use different controllers that have to be accounted for, not to mention the differences between amd and nvidia and how they control memory. Mgpu isn't like sli/cf where the cards work separately together, using just the single vram, mgpu works both cards together as a single card, adding vram, speeds, power etc. Mgpu also makes possible use of any cards, no longer would they need to be identical models, so you could pair a R9 290 with a gtx1060. That alone will be a driver nightmare as it's well known that having amd and nvidia drivers present together is a major source of gpu crashing. So that issue would also have to be worked out and added, as you'll not get nvidia and amd to cooperate on driver compatability otherwise.