I'm surprised it's only a 3x slowdown. But the single-thread performance of native execution (without emulation) is worse on aarch64, which was expected. Imo, a better benchmark would take into account the multithread performance with/without emulation.