Of course this microbenchmark is in the L1 cache so codesize doesn't matter. The article doesn't matter the codesize at all.

I wouldn't be surprised if for cases where this wasn't the only code that you were executing that could sway the overall performance.