Also highlighted in the press release is the fact that this was robust for unconsciously presented happy faces -- it is unusual for positive stimuli to have a stronger effect on most psychological measures. This is usually due to negative stimuli causing higher levels of arousal (as far as I know, though I'd need to confirm this in the literature), so it's really due to a confounding of valence and arousal.
My main question from this article would be how they controlled for low-level visual properties of the stimuli presented. The hope is that their upside-down face experiment was able to account for any possible confounding of face emotion with something like spatial frequency characteristics influencing the participant's subjective judgement of the neutral face, but there will never be a way to fully claim that it was the emotion of the face alone. If you fully control the underlying visual properties of the stimulus, you remove the emotion!