As i understand it, this 'extra' bandwidth is provided by the PLX chip - however, the CPU still only has 16x lanes devoted to PCIe - total
-- the PLX chip is much like USB where there is a *total* amount of bandwidth available. For example, connecting 2 USB 3.0 devices to a hub using a single USB 3.0 port, the maximum bandwidth is the same as 1 USB 3.0 port.
PLX is a PCI-E switch that allows you to connect 2 or more PCIe devices to 1 PCIe port and since the upstream CPU to PLX port is x16x, 16x is also the total bandwidth available to the 2 or more downstream GPUs. As i understand it, using the PLX chip will give similar performance to connecting the GPUs directly to the CPU although the PLX chip has some overhead as well as latency to deal with and you will get less bandwidth (than say X79 MB that has more PCIe 3.0 bandwidth).
i am doing further research and finding conflicting info. i'll contact the EVGA rep tomorrow for clarification .. but this is what i also have found
1 Card: x16
2 Cards: x16 – x16 (Slot 2 & Slot 4)
3 Cards: x8 - x16 - x16
4 Cards: x8 - x16 - x8 - x8
i hope it is the above configuration