I think it was Appel who had a funny proof for the non-existence of the Perfect Compiler, it can be applied to Lisp quite easily.

FWIW, the proof used the halting problem and went like this. Assume there is a an unoptimized program that, when ran, would go through a series of motions but ultimately halt without any useful output or side-effect. Assume there existed a perfect compiler, PC; that compiler would need to be sufficiently smart enough to translate the problem to a single HALT instruction (or "Label X: goto X".) But since the halting of programs can not be proven analytically, it stands to reason that PC does not exist.