So, out of curiosity, I decided to run a test.
Moving a graphics card from a PCIe 2.0 x16 slot to a PCIe 1.1 x4 slot (on the same motherboard, same OS, etc,) resulted in the following drops in performance:
GeForce 9500 GT: approximately 5% drop. (3DMark06 score of 5144 on the x16/2.0, 4868 on x4/1.1)
GeForce GTX 280: approximately a 22% drop. (3DMark06 scores of 15951 and 13101, respectively.)
So a 22% drop, on what is no longer the top-of-the-line card. But... Going from x4/1.1 to x16/2.0 increases bandwith eightfold, so that might not be a super-valid comparison. If I get time later, I'll run the testing with a GTX 295, and compare x16/2.0, x16/1.1, x4/2.0, and x4/1.1. (The motherboard I'm using has an x16 slot and an x4 electrical/x16 physical slot; both of which you can turn 2.0 on or off. I was comparing 'best and worst' before.) This test would be a good one, since x4/2.0 is half the speed of x16/1.1, so I would halve bandwidth every time. (x16/1.1 is also the same bandwidth as x8/2.0, which some motherboards go to in SLI/CrossFire mode; and x4/2.0 is the same bandwidth as x8/1.1, which, again, some motherboards go in to in SLI/CrossFire mode.)