Discriminatory power of stoichiometry-driven protein folding?

There is a saying that when life hands you a lemon, you make lemonade out of it. I found that this has a parallel in molecular modeling: when your test of a putative diagnostic property fails to show discriminatory power, make an invariant out of it. The paper by Mittal, Jayaram, Shenoy and Bawa (1) is a brilliant example of this. Based on three independent observations, comparison of Cα distance distributions, small standard deviation (STD) of the amino-acid propensities and the comparison of amino-acid propensities in structured and unstructured proteins, the paper concluded that the relative frequencies each amino acid occurs in natural proteins is an important contributor to protein stability. This may not necessarily be surprising, but certainly it has not been thought to be the case. Given that these relative frequencies are found to be (more or less) constant, the authors call this set of relative frequencies stoichiometry. I will comment on these observations in the order of strength (as I see it).