Though I think it's more than likely that once people had free access to it and saw what modern conveniences were missing etc the mystique would be lost and it would be kept more as an archaeological asset than as an ongoing thing people wanted to maintain and use.
People would also discover that the architecture is not simple, but grown over a decade while there was rapid development of the basics people take for granted: standards, languages, networking, operating systems, ... In the mid 80s TCP/IP was an expensive add on for Genera. Later it was made a part of the OS. But development stopped before it could add things like IPv6, Unicode, basic security, HTTPS, and other things.
One could add it, but then one had to learn ZetaLisp (70s/80s), Flavors (early 80s), Symbolics Common Lisp (80s/90s), an object-oriented networking stack architecture from the 80s, ...
You would have a second life in a technology stack somewhere between the past and the future, in a parallel world.
Still I also think these kinds of "live editing" environments don't translate well into some modern "best practices" around version control, deployment, versioning, etc. Remembering my experience with LambdaMOO & similar and then later Smalltalk/Squeak and recently with doing stuff with Julia's REPL, etc. They're fascinating and fairly productive, but I am not convinced about this technique's ability to scale.
It may not have the fancy UI, may not have a fancy user experience, may not be an operating system, may have a more primitive Lisp dialect, may not have a good multi-threading story, ... but it might exist and people put a lot of work into it: GNU Emacs and its extensions.
That's how it is...
That plus the lovely keyboards, nice case design, etc. etc.
At some point in time they produced cards for the SUN VMEBus and for the Mac Nubus. Then the only thing left was the keyboard.
So, I agree, the Lisp Machine concept was a combination of Hard- and Software. The emulators of today only provide some of the software parts...
My understanding is Moore's law just ended up making these attempts uneconomical. By the time HW eng and manufacturing is done on the custom components, "orthodox" general MPU capabilities end up leapfrogging it and a well engineered software VM can outperform.
Maybe we could start to see this change again, who knows.
Maybe unikernels running highly tuned VMs on the hypervisor could be this generation's equivalent. Full control over the MMU, tagged pointers all the way down... hmmmmm