i'm willing to bet it's actually computer software which is the cheapest part.. (if talking somethng like the pc having 24cores)
software is also cheaper to develop (when compared to, say, re-re-designing the nmp so it suits your needs)
it's also the part which is most likely to result in noticeable gains when speaking within reason. or- if maxwell 4 goes pure gpgpu for applicable tasks, you're looking at 10-20x faster rendering with the entry level nmp than you would get from, say, a 32core super nmp..
not to mention this is also the scenario which is most likely to play out irl.. or- it's more likely that we'll see new developments in software as opposed to brute force hardware approaches which are in direct relation to the concerns you have.. as i understand them-- and for the record, i'm not trying to argue that your concerns themselves are invalid.. they're completely valid and the entire rendering industry needs great improvements.. what we need at this point is realtime rendering.. the main question regarding that is "how do we get there?"
to me, that question is definitely not "just add some more cpus to a pc".. that wouldn't and couldn't solve the problem.
The difference is that more cpu power will be benefited of right now. And while i totally agree with what you are saying. Most professional software does not rely heavily on gpu yet. And the question is how fast the developers will respond to this move of apple. If 50% of professional software runs opencl in a year i would be surprised.