Assume I have four sets of 1,000 WS.
Set 1: Normalized to the an average damage, D, so the set looks like this:
{D, D, D, D, D, ... , D}
Mean: 1000D/1000 =
D
Set 2: I've "consistently" raised D by 0.05
{D+0.05, D+0.05, D+0.05, ... , D+0.05}
Mean: 1000(D+0.05)/100 = D+0.05 =
0.05/D % increase
Set 3: I've "consistently" raised D by 2% (about the max you're going to get from a "consistent" single piece of gear)
{1.02D, 1.02D, 1.02D, ... , 1.02D}
Mean: 1.02D * 1000 / 1000 = 1.02D -->
2% increase
Set 4: 3% of the time I multiply D by 2.21 and 2% of the time I multiply D by 1.81 (roughly equivalent to what Windbuffet + Ganesha's does)
Now our set consists of
{30 instances of 2.21D, 20 instances of 1.81D, 950 instances of D}
(950(D) + 30*2.21D + 20*1.81D)/1000
Mean=
1.0525D ->
5.25% increase
(2.21 and 1.81 are rough estimates, but they were estimated only allowing QA or TA to proc once, so they're actually ... low estimates).