I agree. The main driving force was size. Implementing a full general purpose CPU takes allot of room using ball logic (even more then using relays, because of gravity). This makes the CPU very tall (see http://brain.wireos.com/?p=2226 for how to implement ROM,RAM and other gates). I originally started by trying to implement a mechanical binary clock, but it became challenging when you need to count to 12 and not to 16. So I used the concepts of a CPU (in particular a program counter, comparisons, and branches). I actually started with the assembly language program and programed it mechanically. However, I believe that this clock can still be used to explain what is inside a CPU (even though its seems confusing). Due to the mechanical nature of the clock, it makes people (not computer scientists) believe that they have the potential of understanding it, so they try. Try that with a silicon blob and a schematics, people loose interest very quickly.

Its a bit more then a counter and an AND gate, since its also needs to check if a value is equal to a number and branch based on that (so it has a one bit program counter). You can also say that a CPU is nothing but switches, or atoms. It is the level of abstractions that give meaning to the physical properties. This particular "computer", could also be reprogrammed by reconfiguring the registers. However, the programing is done by changing the tracks or the flip-flops (see digi-compII or my tutorial on www.instructables.com). For example, the ALU can be changed to preform a different operation by rewiring or switching the T levers to L or I. Therefore, it could be considered as a fixed-program computer, which was how the early computers were programmed, by physically changing the wiring (see ENIAC). However, this "Computer" (and I do use the word very loosely) was not meant to solve-complex differential equations, but to show people how these different components act, and to get them to think about what is inside a CPU.

You are right. The [inc A, if A11 then IP=0 else IP=1] is actually one instruction that was optimised after the first prototype. Originally I had these as separate registers with three other flip flops at the top for the program counter. However, I noticed that I can remove that (which made the clock simpler and more compact) if I combined both instructions together and only use one ball drop for the inc and check. I would love to figure out other ways of explaining this to people who don't even know there is a CPU in a computer. So any suggestions are welcome.

Yea, that my hosting provider. I think it would have been faster if I served the pages from the clock...