According to one paper, "...when dizygotic twin pair correlations are at least one-half the magnitude of monozygotic twin correlations, the assumptions of additivity are met and heritability may be calculated by multiplying by two the difference between Rmz and Rdz (Falconer, 1989). However, in the face of evidence indicating nonadditive gene effects, Rmz should be employed as the index of heritability"\cite{Beatty_2002}(p. 3).
Another way of writing the above statements is as follows:
The use of twin data to test for the violation of definition B of the additivity assumption is also explained as follows by Hill et al. (2008): "...rMZ>rDZ implies that part of the resemblance is due to genetic factors and rMZ>2rDZ implies the importance of non-additive genetic effects. Conversely, rMZ<2rDZ implies that common environmental factors cause some of the observed twin resemblance." These authors also summarized results from a set of twin studies: "On average the MZ correlation is about twice the DZ correlation across a wide range of phenotypes...the simplest explanation of the results is that additive variance explains most of the observed similarity of twins and non-additive variance is generally of small magnitude and cannot explain a large proportion of the genetic variance."
Clearly, Beatty et al. and Hill et al. agree that if rMZ>2rDZ, then definition B is false insofar as non-additive genetic effects seem to be important (whether they are more important than additive genetic effects is another matter). But what if rMZ<2rDZ? Beatty et al. state that in this case definition B would be true, whereas Hill et al. emphasize that this is supposed to indicate the importance of shared environmental factors.
More recently, Polderman et al. (2015) have outlined two other ways to (supposedly) use twin data to test for the relative importance of additive and non-additive genetic variance: "...there are two simple and parsimonious hypotheses that can be tested across traits from estimated correlation coefficients for monozygotic twin pairs (rMZ) and dizygotic twin pairs (rDZ).
  1. The first is that the correlations for the monozygotic and dizygotic twin populations (ρMZ and ρDZ) are the same, consistent with twin resemblance being solely due to non-genetic factors.
  2. The second hypothesis involves a twofold ratio of ρMZ to ρDZ, consistent with twin resemblance being solely due to additive genetic variation."\cite{Polderman_2015}
There is a clear pattern here: if rMZ = 2rDZ (or very close to that), then this is interpreted as evidence that most genetic variance--and indeed most variance overall--is additive in nature. And indeed as noted above this was found by Hill et al., and more recently Polderman et al. reached the same conclusion, at least most of the time: "...across all traits 69% of studies showed a pattern of monozygotic and dizygotic twin correlations consistent with an rMZ that was exactly twice the rDZ". 
The underlying reasoning results from the basic twin equations used to estimate narrow-sense heritability (h2): 
h2+c2+e2=1
rMZ=h2+c2
rDZ=1/2(h2)+c2
h2=2(rMZ-rDZ)
c2=rMZ-h2
Because we defined h2 earlier, we can substitute the definition we obtained to define c2 solely based on the observed twin correlations:
c2=2rDZ-rMZ
So if rMZ is, in fact, exactly equal to 2rDZ, then c2 would be 0.
Hill et al. note that this method is dependent on the validity of the equal environment assumption. Unsurprisingly, they dismiss concerns that this assumption may be false, writing, "Attempts to test this hypothesis have not found any evidence to reject it."