The AMD 2900 was, in fact, how the Xerox Star was built. You'll find quite a bit about that choice in Inventing the Future but certainly not gate-level circuit design. As you implied, they made that decision just before decent microprocessors came along in quantity. Even at the time, there were questions about whether it was the right decision.
Mesa was a virtual machine as well as a language, and its opcodes were selected after seeing what Mesa source code was actually written. The goal was to minimize the code size, and then implement the opcodes in microcode.
After the book came out, my friend Jerry and I wrote a lengthy piece [1] about what Xerox should have done, and this time hindsight is allowed.
“some IP may have been repurposed from the Data General Nova” - the default/bring-up Alto microcode* was/is more or less Nova compatible which is hardly IP IMHO. Different hardware by far and never DG software. I think the Nova was the console controller for the PARC MAXC PDP-10 clone hence their familiarity with the architecture.
*Altos and successor D-machines are soft machines.
Mesa was a virtual machine as well as a language, and its opcodes were selected after seeing what Mesa source code was actually written. The goal was to minimize the code size, and then implement the opcodes in microcode.
After the book came out, my friend Jerry and I wrote a lengthy piece [1] about what Xerox should have done, and this time hindsight is allowed.
[1] https://www.albertcory.io/lets-do-have-hindsight