Probably the most common abuse of statistics in the media is mistaking a correlation for a cause and effect.
Is a study merely observational (a certain group of people show both a certain risk factor and a certain outcome), or is it a much more reliable double-blind study, designed to isolate the relevant factors?
Does moderate alcohol consumption actually contribute to health, or is it simply that healthy people tend to drink moderately? It would be helpful if governments could speak with a united voice and say: even moderate drinking can be dangerous, and there is no particular health benefit from it anyway.
More questions: there are populations that are both obese (have high BMI) and have increased risk factors for diabetes, cardio-vascular disease, etc. Can obesity be identified as a cause of these things? Maybe more to the point: does obesity lead to a higher mortality rate?
There are careful studies that indicate that people with lower BMI may have lower incidence of some specific risk factors, but that doesn't mean they have a lower mortality rate. (A study cited on Drudge today). If anything, skinnier people have a higher mortality rate, but the difference is probably not significant. (There would be no basis to advise people to get fat in order to live longer; but the widespread advice to get skinny in order to live longer is questionable).
So: even if exercise helps you lose weight (and it may or may not), it still won't necessarily reduce your risk of major killers, or extend your life. I would think the benefits of exercise are more along the lines of staying mobile, feeling energetic, avoiding some of the aches and pains of the non-fit (while taking on some new aches and pains), etc. Maybe obesity is like the old joke about vegetarianism (and, er, marriage): it either makes your life last longer, or makes it seem longer.