Thunderbolt 1 is 10Gb/s
Thunderbolt 2 is 20Gb/s
PCIe x8 v2.0 like the Mac Pro you referenced is 32Gb/s
PCIe x16 v2.0 is 64Gb/s
PCIe x16 v3.0 is 128Gb/s
Keep in mind when Intel and Apple discuss Thunderbolt they are talking in Gigabits not GigaBytes like on the PCIe spec pages. To make it simpler I have converted all GigaByte speeds in to Gigabit (GB -> Gb).
Thunderbolt 2 doesn't even reach the same performance as PCIe 2.0 x8 - It is closer to x4 (16Gb/s) and many sites have shown that modern graphics chips when run at PCIe 2.0 x4 speeds greatly diminish in performance. And the problem is compounded when looking at GPGPU workloads like those created by OpenCL and CUDA as those technologies heavily exchange data with the CPU and system memory.
Here is one benchmark showing an AMD HD 5870. This card launched in September 2009. That makes it almost 4 years old. Now take a look at the performance benchmarks:
Image
As you can see by dropping down from PCIe x16 to x4 the average frames per second fell by 13%. Now keep in mind that may not seem like a lot but remember this testing was done with a 4 year old graphics card that is much slower than modern day processors that one may wish to connect over Thunderbolt in an external chassis.
In-fact my own testing with my GTX 780's has confirmed this hypothesis and not with x8 PCIe 2.0 but x16 PCIe 2.0. I actually saw a 300 point increase in the Unigine benchmark just by changing from PCIe 2.0 x16 to 3.0 x16. That resulted in a 7% performance increase in graphics performance.