I am having difficulties to understand the conclusion in bold below, taken from Frankel (1962):
A second limitation of the Cobb-Douglas function appears when it is fitted to historical data. All improved fit generally is obtainable if the exponents: and a are allowed to vary freely instead of being constrained to equal unity. With some sets of data, the resulting sum of the exponents has differed significantly from unity, an outcome that opens the door to economies and diseconomies of scale and that leads to abandonment of the convenient assumption that factors are paid their marginal products.
As I understand things, returns to scale is a technical issue, which might be independent on market conditions. Albeit not a reliable source, the same appears in the Wikipedia entry.
One interesting answer to this might come from the first answer here. However, the author is interested in competition among producers, in terms of the final goods price. Basically, that a natural monopoly (where IRS hold), has an average cost always below the marginal cost, so competitive market produces loses. But what prevents a natural monopoly to pay its factors their marginal products? You can think of a natural monopoly, or any other example of firms with DRS or IRS, in which they operate amid competitive capital and labour markets. I see no a priori reason why market forces would not lead these firms to pay the factors their marginal product.
To sum up: why IRS or DRS mean we should abandon competitive assumptions in the factor markets?