Functions that churn out lambdas - when?

I just wrote my own "repeated" implementation (SICP lecture 3x), and I thought the same could have been done with something of the format (repeated procedure item times), instead of dealing with another layer of abstraction by churning out a procedure that does the actual iteration ((repeated procedure times) item). At this point, choosing between both forms seems like an arbitrary flip of the coin - can anyone on /prog/ enlighten me?

>>11
I saw someone make a CL extension which let you use results from function in the CAR of the S-Expression, this would basically make the code (((repeated times) procedure) item) valid, and still keep it a LISP2. I'm not sure why they explicitly forbid this behaviour when they designed ANSI CL. Personally, I don't mind FUNCALL and APPLY at all, as they provide useful visual cues, but I don't understand their decision to forbid that case.