This argues even more against x86 as it translates to an internal ISA. In effect, the argument becomes that decode-to-risc-cost + risc-execution-cost is always

This argues even more against x86 as it translates to an internal ISA. In effect, the argument becomes that decode-to-risc-cost + risc-execution-cost is always going to be bigger than risc-execution-cost. And this is before you consider the effects of things like looser memory ordering or more registers reducing unnecessary MOVs.