OK, looking at equation (3), this is nearly the equation for a Likelihood Ratio statistic, differing only by a constant (~1.4). Bartlett goes on to do interpretation and whatnot, but he is still committing the equivalent of interpreting the magnitude of a p-value. Under the same assumption of randomness, equation (3) is a random variable is the same way a p-value is a random variable. No, a p-value does not have a “strength of difference” interpretation.
This might have a theoretical “non-centrality” interpretation, related to the statistical power to reject the null hypothesis of “non-centrality equals zero”. Measuring this quantity from data it only a way to restate the p-value of the test.
Other than that, he entirely ignores selection and the efficiency of genetic search, which is darned good, O[n*log(n)].