Are you suspecting any particular analytic error, or are you just expressing an unwillingness to believe in the conclusion without regard for the strength of the evidence?
A drop of more than 25% would be sensational indeed. This is highly implausible. The article even hints at "methodological problems" in previous studies. BTW, medical studies in general are known to be prone to methodological errors.
I work in this area, and I would be inclined to agree with you. The most likely cause of this result is methodological variance. Even if that can be formally excluded (which is unlikely), the next most likely reason (alluded to in the article) is that dementia is one aspect of ageing, and that if people are healthier for longer then the effects of ageing are less pronounced in the younger elderly. Thus it would have no impact on the eventual likelihood of dementia for an individual, just that it might occur later than it otherwise would.
