I agree that current hardware could be improved, but at the cost of user development, I say that because as I'm sure you have noticed most of your hobby stuff is going to be developed on something like the ATmega. Why? because it is familiar and easy to work with.
For something like an advanced autopilot perhaps with computer vision you are going to need lower level hardware making near instant decisions reguarding stabilization via PID loops. But for CV and map data you are generally going to need something much higher level to do much on the user-friendly side of things.
I suppose if you needed these two things at the cost of ease of development you could sue something like a multi-core mobile phone processor, and some lower level hardware for PIDs. However to do this you are likely going to need to be very skilled in hardware and software development so perhaps this is limited to larger companies with extra money to spend on R&D. The improvements are not going to be very important just yet, and you will likely lose much on the community driven opensource side of development. And of course price.
What tangible improvement are you really getting from faster loops over that of a traditional flight controller?
