At one point I was seriously fascinated by the lambda calculus and particularly the S and K combinators - two very simple functions that you can build any computation from:
Having a system that implemented recursion using the Y combinator in terms of S & K and have it actually compute stuff was rather amusing. I'd be fascinated to see what it would look like using just U.
Actually, one of the many things I learned from doing a final year project for my CS degree on "benchmarking" different sets of combinators was how to write a garbage collector - I was doing this in '87/'88 on a Unix mini (HLH Orion) and memory was tight and my trees were comparatively huge!
It's worth noting, BTW, in case anyone is still following this thread, that what is going on here in all of these single-instruction languages is not particularly interesting. You're just encoding the program semantics in the arguments, or in the structure of the tree, or the length of the program, or something like that, rather than in the instructions.
http://en.wikipedia.org/wiki/SKI_combinator_calculus
More recently I found out about the U combinator which is a single combinator that can be used to define S & K:
http://en.wikipedia.org/wiki/Iota_and_Jot
Having a system that implemented recursion using the Y combinator in terms of S & K and have it actually compute stuff was rather amusing. I'd be fascinated to see what it would look like using just U.