The author appears to be benchmarking the specific operations that bottleneck their json parsing library when running on an Intel chip, which seems reasonable, on its face. It can fail if the library is limited by a different set of operations, on the different machine. But that is unlikely if the specific operations tested are slower.