Yep, this is exactly what I was wondering. Opcode caching is a key part of production PHP environments - in this case, since the code isn't changing, they should even disable the "change check" (apc.stat=0) as one would do in a production environment.

And if this is the case with their configuration of PHP, it makes me wonder what other platforms are not configured for production in this benchmark :)

The relation of PHP to everything else in this benchmark is so unusual among benchmarks, that I strongly suspect a failure of parity with other configurations. Hopefully that is unintentional.