Being a one application system is in conflict with continues to sell (well).The Xserve sales numbers were a fraction of the Mac Pro sales, ...
If the Mac Pro dips under 50,000 units a year, it might be in trouble. I think Apple is perfectly happy to have the Mac Pro just be a great FCP machine as long as it continues to sell.
FCP runs on other Macs. Over time some users are going to bleed to those other systems. Similarly, there are other apps folks in this workstation class will want to run. The "max market segmentation" of combing a narrow platform app with a even narrowing hardware platform most likely will shrink over time.
With that "FCP is good enough" strategy, the Mac Pro would be on very similar same path as the Xserve ( The Mac only server management tools were suppose to save it... not). What folks largely needed was a server that happened to have some Mac specific services on it; not the other way around.
The Mac Pro is still a 'truck' . It may be a fixed bed pick-up truck (versus a Big rig semi tractor trailer), but if don't have computational horsepower problems then don't really need it. Only covering a large enough breath of computational problem solving is the Mac Pro going to survive long term. OpenCL 2 and AVX2 will open the breath substantially.
Letting the Mac Pro go 'stale' on hardware isn't going to help them hit 60K/yr run rates. They are protected a bit when looking at the moving year average right now because of the 2014 launch bubble driven by pent up demand. The problem Apple will run into in upcoming months is there enough volume to keep the contract manufacturing going at a reasonable (to the contractor) rate.
Apple could go into hibernation until build another > 60K demand bubble again. Just wait 2-3 generations of GPU/CPU technology and then only compare to the system from 2-3 years ago and ... yeah can get "ooh look it is faster" bar graphs pretty easily.
The question is whether folks in those "in between" year(s) are going to wait around while Apple "innovation" amounts to just waiting longer periods of time between upgrades. Apple could just simply follow the even Xeon E5 sequence ( v2 , v4 , v6 ) ,but that strategy is adept at milking a shrinking cash cow pool for 2-3 more iterations than a healthy ecosystem.
The flaw in waiting for any one tech ( CPU , GPU , thunderbolt) to get to some specific version to do a release is that they are on different cycles that won't necessarily align going forward. The computational horsepower of this new design isn't solely dependent upon any one of those. Waiting for a trifecta of all them to update at once is like herding cats.