In common lisp, a for loop is reduced to a goto internally via a macro which manipulates the syntax tree. In C, a for loop is internally reduced to a goto after the code is transformed into a syntax tree (manipulated by a non-standard and ad-hoc macro implementation). The difference here is that C must first make the syntax tree before it reduces the tree while CL reduces it immediately (since lisp code is already a tree).

All languages must reduce to a limited ISA of primitives (or conversely, all HLL are built on small ISAs). Lisp does a lot of this conversion early on (because it can) while C does this much later (because it can't do it early). You're simply quibbling about how soon to do this conversion. LLVM does the same thing as lisp in that it reduces a seemingly large language into a small set of constructs (and then works to optimize this small core language). I would argue that a smaller core language is better because it is easier to reason about and easier to optimize.

As to embedded, iRobot corporation created a lisp variant called L (based on a subset of common lisp) to use on their embedded processors via a bytecode VM.

http://www.cs.cmu.edu/~chuck/pubpg/luv95.pdf

Check out shen lisp (link below). It has (to my knowledge) the world's only Turing complete static typing system. Racket also has a static variant. Common lisp can be made into the equivalent of statically typed if you (declare) all variables and tell it to fully optimize.

http://shenlanguage.org/