Well well I just did some testing on test live servers and my findings are completely different than yours.
What I did was, on both my conqueror and my guardian, i reseted the feat trees. Unequipped all my gear and weapons. Unequipped all AAs (passive ones I can't do anything about but they're irrelevant anyways).
I got a 20 level white 1 handed blunt which had a base dps range of 19-25. I popped a strawman and made lots of white hits on that totally naked toon. Then I got enough feats in the general tree and feated sustained rage to all 5 points for 15% melee weapon damage. I repeated the first step and made lots of white hits on a strawman with same weapon.
The only changing parameter for my damage numbers was that in the first phase I didn't have sustained rage, and second phase I had sustained rage. I repeated these phases 3 times, resetting feats each time and starting over. For both classes:
In the first phase my white hits on the target ranged from 31-40. The median value was around 35 and mode was around 33. I had 1-2 hits at 40.
In the second phase my white hits on the target ranged from 31-42. The median value was around 37 and mode was around 38. I had 4-5 hits at 40 or above 40.
I completely ignored crits, misses, half-hits.
My observation shows that even with a weapon that has very low base damage, sustained rage provides a visible increase in the damage numbers.
This seems like a very simple test and I didn't bother to write down every single damage value, I just made about 30 white hits in every phase and looked at the numbers. However, I'm quite sure that this simple test is way more accurate than parsing a rotation of combos over a strawman for 3 minutes. In a strawman test there are SO MANY variables (human factor, gear, combos, crits, misses, debuffs, crits hitting at debuff/buff periods, etc...) Therefore looking at a strawman test the benefit of a 15% melee weapon damage increase can easily be overlooked. Try to come up with more accurate tests if you can but do not say that the feat is bugged/useless. Make sure your test eliminates all other variables before you reach a conclusion.
p.s. here is the build i use on my SnS: http://goo.gl/VW6dL