It isn't representative of how processors work, unlike imperative programming. Even if the theory is there, the machine itself is imperative (most of the time), so there's a level of abstraction needed.
Though the theory could have started like that, the mechanisms are totally separated from the physical reality. In a field dominated by engineers (especially in a time where cycle counts mattered much more), that's what's going to happen.
> There were honest to god LISP machines
As LISP-flavored stack machines, yes [0]. Alas that Linear Lisp [1] never received a hardware implementation [2].
To those exploring high level language architectures, forget not also the noble SCHEME-79 chip [3]: a Von Neumannian register machine, but one which executes a SCHEME interpreter microcoded in LISP.
[0] https://en.wikipedia.org/wiki/Lisp_machine#Technical_overvie...
[1] http://home.pipeline.com/~hbaker1/LinearLisp.html
Since when functional programming is "radical"? It's the essence of computing.
What we're seeing are other languages going back to the basics, after a detour in the "imperative is easy to learn" and "OO is reusable" lands.