Most single-locus molecular approaches to species delimitation available to date have been designed and tested on data sets comprising at least tens of species, whereas the opposite case (species-poor data sets for which the hypothesis that all individuals are conspecific cannot by rejected beforehand) has rarely been the focus of such attempts. Here we compare the performance of barcode gap detection, haplowebs and generalized mixed Yule-coalescent (GMYC) models to delineate chimpanzees and bonobos using nuclear sequence markers, then apply these single-locus species delimitation methods to data sets of one, three, or six species simulated under a wide range of population sizes, speciation rates, mutation rates and sampling efforts. Our results show that barcode gap detection and GMYC models are unable to delineate species properly in data sets composed of one or two species, two situations in which haplowebs outperform them. For data sets composed of three or six species, bGMYC and haplowebs outperform the single-threshold and multiple-threshold versions of GMYC, whereas a clear barcode gap is only observed when population sizes and speciation rates are both small. The latter conditions represent a "sweet spot" for molecular taxonomy where all the single-locus approaches tested work well; however, the performance of these methods decreases strongly when population sizes and speciation rates are high, suggesting that multilocus approaches may be necessary to tackle such cases.