Manufacturers generally don't OC their own hardware when creating these types of comparative benchmarks, as many would interpret it as an official sanction of the practice and giving people ideas that a particular OC level is 'guaranteed'. I can't recall a single instance of it off the top of my head. Rather, traditionally, it would mean selecting the computer hardware, screen resolution, game detail settings and level/scenery/benchmark run that gives the most favorable figures compared to the competitor's offerings.
You might also suspect that GPU control panel settings could be tailored to widen the gap as well. I suggest not paying attention to hardware vendors' numbers and instead looking at those of "independent" reviewers instead. While not perfect, they're going to be less flawed than what - for instance - AMD themselves will come up with.
