I'm talking about the overhead post-translation for Rosetta. Translating optimized code from one architecture to another is difficult, which is why there is a performance hit.