Nice to see your thoughts on this. A few further thoughts/points of clarification.

The reason that alleles that are under-dispersed are hard to detect is that for low differentiation there are many loci that have only small changes in frequency between populations. If we look at the distribution of Fst across loci, and Fst is not very large, this distribution is L shaped and there are many loci with near zero Fst. This means that a test that looks for under-dispersed loci is under-powered, when populations are only recently diverged, and tests of extreme differentiation are much more powerful (e.g. see here). Over longer time-scales we could hope to see loci that maintained polymorphism, and indeed looking for trans-specific polymorphism is one test of balancing selection (and conversely tests of extreme differentiation are very underpowered as many loci are fixed differences).

That said, the idea that balancing selection leads to strongly under-dispersed allele frequencies among populations has long seemed strange to me. Strong heterozygote advantage (or other simple negative frequency dependent models) would keep the allele frequencies relatively constant, if the relative fitnesses of the genotypes stayed constant. However, it seems likely that these fitnesses will vary across environments. Also under more general models of balancing selection, temporal or spatially selection, we are invoking variation across conditions to maintain polymorphism and so we shouldn’t expect frequencies to stay constant.

We’ve developed a test for loci that strongly differ from a covariance matrix of allele frequencies (see here), that is along the lines of the one suggested by Joe but that has some nice features that make it a close analog of Fst. Once again I think this is better at detecting overly differentiated loci than balanced loci [but our tests of environmental correlations may well be a good way of finding spatial balanced polymorphisms].

Anyway great to see your thoughts on this, and hope this helps.

Graham