For some perspective, there's a 2005 paper in PLoS Medicine by John Ioannidis in which he predicted that 80 percent of non-randomized studies turn out to be wrong, for a variety of reasons. Being published in Science indicates novelty of the results, not accuracy. Science mag publications usually generate other research (enormous amounts) that gets closer to the real answer. The scrutiny applied to this XMRV paper is unusual however. I suspect if every publication had this much attention, you would see many more inaccuracies and retractions.