Aggregating those figures is not generally wrong. However, to fully saturate all this bandwidth, the GPU workload must read and write roughly the same amount of data at the same time. And, of course, most of that data has to be cached. The GDDR5-setup is more flexible.
For me personally, there are two reasons why I might prefer the Iris Pro: First, having more power in light workloads, especially driving the UI in scaled mode. I saw that the 15' rMBP's HD4000 can struggle with UI effects while running in scaled mode, and I am not sure that the HD4600 will be that much better. Second, the OpenCL performance seems to be very nice and would be a plus in FCPX.
Having a 760m would make the decision much easier. But I don't think that this will happen.
Excuse me please, what is that "openCL", that is mentioned so often in this thread?