Also repeatedly zeroing the same memory can have different performance characteristics than zeroing different memory each time. Haven't checked what the benchmark does though.