I agree. A counterargument is that you're inherently taking advantage of the processor(s) and the operating system, not to mention, ha ha, every single component and factor that you did not personally create, whether or not it is computational in nature...

Another route, my present hobby and something that I know many have done before: Take the Lisp program on page 10 of the Lisp 1.5 User Manual and translate it into C. Write a simple tokenizer/parser (read function) in C and translate that into Lisp. Once the C program can run the equivalent Lisp program, we're on our way.

At the outset I have no garbage collection. I just expect the memory space to fill up pretty soon. I could have `cons` compute a hash for each cell to eliminate common subexpressions, but for a while I suppose I won't bother at all. The C program has a loop to handle evcon (did I write cons? oops), eval, and apply as a trio of mutually tail-recursive operations. That's an obvious place to put a stop-the-world mark-and-sweep operation or some such thing.

To get the C tools out of the loop eventually, hand-disassemble the binary program (C compiler output) just to see what's there. Then write a Lisp program (compiler) that translates the Lisp 1.5 User Manual program into something pretty similar.