You could also write a meta-interpreter in CL for the limited subset defined in the original paper in about 5 minutes and just have people use the interpreter. From there, sure, everybody could recreate CLOS if they really wanted to, but it would take them a while.
BTW, why the requirement for things to be so secure? If it's just a contest for fun, can you not rely on people to adhere to the rules without having to jump through hoops to prevent cheating? Or maybe I'm misunderstanding the point of the exercise. I mean, if there's $1M hanging on this, or something, then yea, I suppose you need to take steps to prevent cheating. If it's only for your own personal self-worth, you'd have to have pretty low self-worth to cheat your way toward self-worth.