It depends entirely on how parallelized the code is. This is more than whether the app is multi-threaded. The time-consuming computational part must (a) algorithmically lend itself to parallelization, and (b) the programmer must effectively harness multiple threads for that task. Neither one is a given.
Algorithms break down into serial and parallel portions. It has long been known that even a small % of serialized code will limit available multi-threaded speedup. This is called Amdahl's Law:
http://en.wikipedia.org/wiki/Amdahl's_law
You can see from the graph if the code is 90% parallel (only 10% serial), that a 32-core machine will only be 8 times faster than a single core machine.
However additional cores can help beyond a single program. They are available for running multiple independent programs. A 12-core Mac Pro could more effectively do concurrent activities without bogging down: render, editing, download, transcode, etc.
To the user this would not appear as a single activity being magically faster (than an iMac). Rather the performance would degrade slower under greater multi-program loads. The machine would feel like it has "lots of torque" and doesn't bog down under heavy loads.
Of course if your independent tasks are so isolated they could be done on two different iMacs, you'd have to ask whether two iMacs are better than a single Mac Pro (possibly the same total price, depending on configuration).
Specific narrow tasks like 3D CAD/CAM could be greatly accelerated by the Mac Pro GPU. If you need that it would be worth the price.
4k video editing pushes around lots of data. Also newer high-compression codecs like H.265 (needed for 4K capture) require lots of CPU to handle. A Mac Pro could work well for this.
If Apple figures out how to fully harness the Mac Pro GPU for video transcoding, that by itself could be worth the price -- if that activity is your frequent job. It conceivably could be 5x or 10x faster than the fastest iMac.
