On that subject, Richard Gabriel [1] writes about his experiences as a founder of Lucid, which produced Common Lisp for regular Unix workstations, in the "Into the Ground: Lisp" chapter of his book, "Patterns of Software", which is available as a free PDF from his web site [2].
(Lucid's pivot to developing a C++ environment is covered in the "Into the Ground: C++" chapter).
There's some interesting history there (and the rest of the book is probably worth reading as well for a variety of reasons).
Lucid took the money they earned and invested it into some ill-fated and ill-designed C++ environment. Lisp competitors from that time, Franz Inc. and LispWorks are still in business.
I couldn't afford a Lisp Machine, I just used Franz Lisp on an Atari ST.
Basically they couldn't amortize the NRE for their custom hardware and later most especially chips (from memory, first, a chipset that spread the CPU across several chips, rather like the one Western Digital did that among other things enabled the LSI-11, then of course an all in one chip) across the huge number of units that Intel and Motorola sold. They also canceled their RISC rethink of their basic low level architecture on the day it was supposed to tape out; don't know if it had a low enough gate count to be like the first SPARC processor, which was implemented on 2 20,000 gate arrays, one for the CPU and one for the floating point unit (gate arrays are all alike until a few layers of metal are put on top).
So soon enough you could run a full Common Lisp, almost certainly without as much run-time error checking, faster on cheap commodity hardware than on a Lisp Machine.
Something like that seems to have happened to Azul Systems, which apparently isn't developing any new hardware, but is selling their version of the HotSpot JVM to run on x86_64 hardware. A prior generation of their pauseless GC (vs. 1 second per GiB, a big deal if your heap is 100s of GiB) required a software write or read barrier that cost ~20% of performance (all this from memory). It's likely that soon enough, even if it perhaps ran slower than their custom hardware, it was a lot cheaper to run it on commodity Intel/AMD hardware.
The base papers are:
Pauseless GC, uses a read barrier instruction in their custom 64 bit RISC chips: https://www.usenix.org/legacy/events/vee05/full_papers/p46-c...
And the newer one that they're using in that old hardware and the software on commodity hardware Zing JVM, the Continuously Concurrent Compacting Collector (C4), which I have not studied (the paper was published 2 weeks after the Joplin tornado trashed my apartment and rather disrupted my life): http://www.azulsystems.com/sites/default/files/images/c4_pap...
It's possible they figured out how to minimize or eliminate the penalties of the original software read barrier they applied to the Pauseless system in C4 (or perhaps in relation to their custom vs. newer commodity hardware); I just did a quick skim of the relevant part of the C4 paper and a few keywords and couldn't tell.
This is all great stuff that I hope to get back to soon....
You might've been able to buy a Xerox workstation for, say, $15K-$20K, while a VAX box was over $100K. BUT for a production system, the VAX could run multiple Lisp processes at the same time. (I'm guessing at costs here, it's been too long.)
Also, Richard Gabriel did a bunch of Common Lisp benchmarks, might help to look for them. (Great fellow.)
The real threat to Symbolics et al, circa 1987, was things like Gold Hill Common Lisp running on, say, a IBM PC/AT with a 286 chip and maybe a meg of memory. At, perhaps, $3,000. It ran pretty fast, the Gold Hill tech folks were very good. But GHCL had an unsophisticated development environment compared to Symbolics / Xerox.
As excellent as the workstation environments were for development, a market demands that you eventually deliver a cost effective product into customers' hands. (Or is that too much old-think?)
As part of out work, we evaluated and benchmarked Xerox Interlisp machines, Symbolics systems, VAXen, later Gold Hill etc. to find a cost-effective delivery platform. We even eventually funded the development of a delivery-focused subset of Common Lisp.
One aspect that Symbolics didn't seem to understand back then was cost of entry and deployment: the Xerox D-machines were (IIRC) around 1/3 the cost of the Symbolics. Perhaps not as speedy, but adequate for our day-to-day development work as well as for the end customer's needs.
Symbolics had great development systems, but the delivery answers were late in coming; too late to help us.
There's lots more to be said about the late 80s collapse of AI (ES) applications and expectations, but the margins here are too small to contain it....