if you already made two cards work, four cards is going to be a lot easier (than designing from scratch again). imagine optimization problems though. at a certain point, there's just no more optimizations to be had (for GAMES). i'd rather them spend all their engineering power on getting new generation cards out the door than this, wouldn't you? no, im not bashing SLI. there's no doubt you get good performance but you'll pay for it. but I would never SLI together FOUR current-gen cards. the most I'll ever do is SLI two this generation, then SLI two next generation because that makes a lot more sense to me. plus it would be more efficient, economical, and usually more future-proof.
one practical usage is workstations, but that should have been developed a while ago (like the Evans and Sutherland implementation) because the current speed of rendering is never enough.