No matter how good it gets, binary recompilation will always be slower than native code. And when the host CPU is already a very slow model like Qualcomm ARM, those effects stack.