Non-random results are a common tool for identifying fraud, not just in science papers but in ordinary crime. In the book The Design Inference the story is told of a Democrat operative who was supposed to randomly choose which candidate was listed first on a state ballot in elections (because it is known that whomever is first gains a significant advantage). In 39 of 42 ballots they put the Democrat first. This was enough to earn them a conviction as it was outside the bounds of reasonable probability.
In another case, a criminal was caught because a search of an insurance claims database showed he kept claiming to be injured in restaurants and shaking down the owners for a quick payout. "I slipped on a mint jelly packet in the bathroom" only works so many times - particularly when some of the restaurants didn't even carry mint jelly.