It has to have more latency than existing dual GPU setups. Since GPU performance is not synchronous, the data has to be split, sent to GPUs, processed, sent back to chipset and reassembled before going BACK to the video card and going out to monitor.
At least I think..I have been known to be way off base