I have been doing a fair amount of research along these lines, as well, lately. Based on what I’ve read so far, the only real reason to go with Quadro/Tesla is if you need the big RAM that some higher end cards provide. These cards are a fantastic amount of money, though.
I’ll toss this out , but am open to correction, as this isn’t personal experience, yet :
The Quadro (and Tesla) cards seem to be regarded as QCed better than a consumer-market GeForce, draw less power, use ECC memory, have double-precision floating point cores. Blender/Cycles uses single precision math, so the DP capability is an expensive, unused option. I watched a few YouTube comparison using non-Blender rendering software, of Quadro VS GeForce-4xx, and it is nice, but I believe the new GeForce 500-series cards are the best nVidia FLOPS-per-dollar solution for Blender.
Cards which are PCIe 2.0 compliant apparently cope well with backwards compatibility to PCIe 1.xxx. PCIe >2.0 (2.1, 2.1.x) seems to be hit and miss. All the nVidia hardware I’ve looked at has stayed at PCIe 2.0, so I am assuming there will be no issues with any of my PCIe 1.0 lanes when I start to populate them. Can’t comment on ATI.
As @Writer’s Block posted, nVidia is supposed to be hitting the market with new GeForce 600-series GPUs very soon. Denser architecture, less power consumption, more compute power and probably a lot more (faster) RAM. Lots of reasons to make do, for now, learn a few more techniques or something. Even if they are out of y/our price range, the 500-series cards should drop in price because of the new hotness. I’m hanging tough until I see what they’re offering.