I think it's the opposite actually: the twins had similar exercise routines while growing up but diverged in recent years.
The dissimilarities in their exercise routines had mostly begun within the past three years
The dissimilarities in their exercise routines had mostly begun within the past three years
In this case, though, we can use a single individual with "before and after" analysis to show that exercise improves those parameters. No identical twin is needed as a "control group of 1"; your former self, say, six months ago, is far better than an identical twin in most respects except for one: you, six months ago, is half a year younger. That is insignificant, and even a good thing for the experiment if the experiment confirms improvements in parameters that otherwise decline with age. The results can also be shown in corrected form, using estimates of age-related decline from a good body of data, fit to the duration of the experiment.
It is a bit of a shortcut of the researchers to just find such identical twins with a recent divergence in exercise habits. I mean, you find the twins, run some tests, and there is your data! You don't have actually conduct a experiment that takes months or years, where you face the risk that subjects will drop out.
You can also cherry-pick the twins whose data agrees with your results. Signs of fitness, like lower body fat, are obvious and an obvious bias. If the researchers found twins who didn't show such signs (yet claimed to have had different exercise habits for a few years), I suspect the researchers would be inclined to drop those twins from the data, rationalizing that choice by a justified disbelief.