The incentive is how much your utility increases - between state B and state A, that's u(B)-u(A) = u'(h)(B-A). Take limits and you get to u'(x) for marginal estimates.

Inequality is most definitely NOT defined by ratios of a utility function. The convexity of the OP was in the utility function; he seemed to believe that convexity of reward (and hence of utility, assuming u(reward) is monotonic) would only increase the reward for those at the top.

Maybe you should have made clear in your post, then, that u is utility, not income. If u is utility and v is income, then v'(x)/v(x) ~ u'(x), more or less. Most measures of inequality involve dollar ratios, which seem to correspond to differences in utility, or at least to self-reported happiness.