Molecular phylogenetics has entered a new era in which species trees are estimated from a collection of gene trees using methods that accommodate their heterogeneity and discordance with the species tree. Empirical evaluation of species trees is necessary to assess the performance (i.e., accuracy and precision) of these methods with real data, which consist of gene genealogies likely shaped by different historical and demographic processes. We analyzed 20 loci for 16 species of the South American lizards of the Liolaemus darwinii species group and reconstructed a species tree with *BEAST, then compared the performance of this method under different sampling strategies of loci, individuals, and sequence lengths. We found an increase in the accuracy and precision of species trees with the number of loci, but for any number of loci, accuracy decreased when using only one individual per species or 25% of the full sequence length. In addition, locus 'informativeness' was an important factor in the accuracy/precision of species trees when using a few loci, but it became increasingly irrelevant with additional loci. Our empirical results combined with previous simulation studies suggest that there is an optimal range of sampling effort of loci, individuals, and sequence lengths for a given speciation history and information content of the data. Future studies should be directed towards further assessment of other factors that can impact performance of species trees, including gene flow, data 'informativeness', tree shape, missing data, and uncertain species boundaries.