In it, several people were given the NYT crossword puzzle to do several days before it was published. They averaged how far and how long it took the subjects to work the crossword puzzle.
Then, AFTER the puzzle had been published (and people around the country had worked it), they tested a new group of people (who hadn't worked the puzzle). THOSE people did remarkably better, on average, than those who worked the puzzle BEFORE it was published.
Maybe the second group was just plain better at crosswords, I dunno, but it was a strange experiment.
There are probably more reasons for this. Something as strange as testing at different times of the day could influence the results.