The exercise/akrasia correlation found here was pretty much nonexistent, and you'd expect lack of akrasia to cause exercise, so if anything this seems to me to be evidence for exercise being harmful.

ETA: On second thought, I'm assuming (p=.3, c=.02) is a typo for (c=.3, p=.02). Then it's not clear to me, after accounting for the above effect, which way the evidence goes.