I don't see why it matters if everyone who's overweight loses 2 points or if a fifth of them each lose 10 points.
Assume that BMI is actually useful. Then, assume everyone below median dropped 2 BMI points, and everyone above median stayed the same. This would lower the mean without having positive health effects (it might even have negative health effects). If you flip it, so the above-median lose weight, it'd have more positive health effects despite yielding the same mean. The overall public health benefit is simply not determined by the average effect, because 1 unit of BMI != 1 unit of harm.
If most of the harm comes from the extreme end (e.g. BMI 35+), then having the top 10% drop 10 points would alleviate a lot more harm than having everyone drop 1 point, even if the effect on average is identical. On the other hand, change among the people who are most severely overweight seems less likely to actually happen than slight change across the whole population.
I agree that it matters who is losing the weight (that is, where in the BMI distribution they are). That's the point I was trying to make in my 2nd paragraph. But it doesn't matter what increment they're losing the weight in. It makes almost no difference if a group of people all lose 2 points of BMI or if a randomly selected half of that group all lose 4 points.
If I'm reading them correctly, Paul Campos suggested that a 2 point drop in BMI was too small to make a difference, and Jonathan Graehl responded that a 2 point average drop in BMI could be big ...
Related To: The Unfinished Mystery of the Shangri-La Diet and Missed Distinctions
Megan McArdles blogs an interview with Paul Campos, author of The Obesity Myth. I'll let anyone who is interest read the whole thing, but here's some interesting excerpts:
and