Compiling to bytecode is still compilation. An interpreter would operate directly on the S-expressions. Obviously, this makes some constructs (such as TAGBODY) very inefficient.