I didn't get it either, but video editing software is designed to load rendering to the dGPU when it's present. Some effects are even dependent on CUDA in some software. I'm not a pro by any means but I've been editing video/audio for about 15 years starting with VHS tapes. I used FCP7 on a late 2009 mac mini for a while and it was fine. The rendering took a while but I'm not in production so it didn't matter. Then trying to use FCPX and Motion I hit the wall. The whole process became extremely painful. When I had a high-end rMBP w/650m the difference was obviously impressive but it was overkill. I had 3 different Motion templates rendering at the same time and it could take it.
Yes, sure, modern GPUs are computational powerhouses and can assist you tremendously in these tasks, but these facts are not really helping in the whole CPU vs. GPU debate. First of all, the previous Nvidia GPUs (650M/750M) are quite bad at those tasks (due to Nvidia optimising the cards more for gaming performance) compared to some older cards (Maxwell is quite fast again). Secondly, the Iris Pro is every as bit capable at processing data as a midrange GPU and is actually much FASTER when random data access is involved (as OpenCL benchmarks clearly show). The only scenario where the Iris Pro comes short is when memory size/bandwidth becomes the limiting factor - such as rendering very large videos with simple filters (but do you really do that?).
Bottomline is - if you try the MBP with Iris Pro, I am sure you will find it just as fast as that laptop with the 675m.