Spin the Cylinder?
The cleverness of the 2013 Mac Pro's design (inadvertently?) became more important than performance headroom.
Federighi is careful to note that, while Apple set out do so something new and different with the Mac Pro, “we didn’t start with a shape and say, ‘well, here’s the fastest machine we can put in that box.’ We actually started with a target for performance and came up with what I think was a very clever design of that thermal core and thermal architecture to accommodate what we thought was the right power architecture.”
This completely undermines my thesis. Naturally, I don't believe a word of it. :-)
Edit: My friends helpfully point out that it doesn't undermine my thesis. The combination of "a target for performance" and "a very clever design" is precisely the problem I'm highlighting, where the cleverness of the design (inadvertently?) becomes more important than performance headroom.
In particular, the expectation of highly parallel GPU workloads expected a wholesale change in direction across high-performance computing, where Apple was not already a leader shaping implementation strategies. It hints at hubris.