CUDA performance when 16x card cut to 8x?

I have no idea what actually impacts CUDA performance, so I need to ask.

My GTX 285 is currently in a 16x PCI-E slot and doing OK. Badaboom (only CUDA app I use so far) has its highs and lows as far as frames-per-second go, but 90FPS is about the lowest I’ve seen. Even that low, about 3 times as fast as I can get out of a CPU-only convert.

I have some lame SATA ports that aren’t hot-swappable, and want to install more drives in SATA trays. To do that, I’d need to get a new SATA adapter, and one reasonably-priced Adaptec model is a 4x card that’d have to go in my second video card slot. Problem is, this motherboard will cut both video slots to 8x if both are occupied.

So my question is, what kind of performance hit will CUDA take when the video card bus is running “slower” than before? If there’s a way to simulate this kind of speed drop with some utility I’m willing to try it.

The $150 the Adaptec card will cost could go toward a motherboard with more/better SATA ports and no 8x limitation, so knowing what I’m in for up-front is important. I’m not interested in crippling CUDA just before I get around to learning how to program with it.