They very much believed that their optimized interpreter would be faster and better than JITing. And it is optimized -- they took into account cache locality, A

They very much believed that their optimized interpreter would be faster and better than JITing. And it is optimized -- they took into account cache locality, ARM instruction size, etc. You can read the design doc. They spent a lot of time on it.