I had used NJ method for phylogenesis with allele frequency of two loci among 45 human populations but a reviewer had questioned the reliability of NJ method and suggested me to try with Maximum parsimony in POY.
I don't know about POY, but you could probably try using TNT for an MP analysis, using continuous character coding. Similarly the program CONTML (in Phylip package by Felsenstein) includes an option for ML analysis of allele frequencies. May be of use.
You certainly can use POY to reconstruct phylogenies using allele frequencies with MP. The most notable difference with other software is that POY uses direct optimization.