I stand corrected on that point then. Thanks for the info!

So it looks like they cherry-picked the 6 subjects who did (by chance) worse on the memory task than the other tasks. Then regression to the mean happened, and voila! Some positive results.