Yea, that's a good question. All I know is that even going from PCI-E 8x -> 16x, there wasn't a doubling of performance either or vice versa 1/2 the performance. You can take a look at this analysis here for PCI-E scaling:I've been in contact with John. Nice guy and quick response times.
So I guess my question would be the following: How can you get the raw computation power from those monsters *if* for arguments sake the bandwidth is capped? Doesn't every application rely on the amount of bandwidth available?
I think it really depends on what you're using your system for but for what I use it for it's mainly floating point analysis and matrix manipulation and it's more CPU bound. i.e CPU can't pre-process the data fast enough to feed the GPU.