Much of the memory in a Lisp program is in the form of CONS cells, so the MIT LISP machines had a compact way of encoding this. They also used tagged memory to be able to handle the different kinds of data at runtime. They inherited a very stack-oriented execution model from the PDP-10 implementation of LISP. And they implemented very complex instructions using microcode.

The Symbolics people refined this approach while the LMI people kept the original design until nearly the end when they tried to do a RISC+tags:

http://fare.tunes.org/tmp/emergent/kmachine.htm