You shouldn't dismiss so lightly. The cited document answers the question of the first target, which was the National 32000. It ran on a Sequent multiprocessor, a machine we were thinking about using for other work.

So no, it was not created on Research Unix, or at least not for it. We wanted a faster compiler for a number of reasons, and Ken had ideas about that. Because of its architecture, it was easy to port to other RISC architectures, such as the MIPS and SPARC, although both those machines had wrinkles that needed ironing out.

-rob