If you refer to the *MainEvaluator *and rule based approach inside Mathematic plus that there is no real function call semantics... no. Performance comparisons must be absolute, despite of a (sometimes) nebulous execution model Mathematica has.
Are you saying, that if you'd use a similar execution model that Mathematica has you gain more speed? You're able to change the execution model of a llvm IR code generator? How you gonna do this? I don't think that you can change the code to gain more speed, because you involve a special execution model, but I do believe that you can better Julia code that I've done.
