Eh? How can he take a binary from iOS and run it on an Intel machine?
There's a misunderstanding: I was talking of Poole (the GB author) results. He can obviously recompile his own application. What he did was to run on the same iOS device the mobile and the desktop datasets. This is what gave the 1% difference.
It's possible that doing the reverse, that is running the mobile datasets on a desktop machine with an x86 CPU would give a different variation. This is what Piednoel claims. But as I wrote I expect some more solid evidence than just a tweet :-)