Genomic data provide unprecedented power for species delimitation. However, current implementations are still time and resource consuming. In addition, bioinformatic processing is contentious and its impact on downstream analyses is insufficiently understood. Here we employ ddRAD sequencing and a thorough sampling for species delimitation in Zodarion styliferum, a widespread Iberian ant-eating spider. We explore the influence of the loci filtering strategy on the downstream phylogenetic analyses, genomic clustering and coalescent species delimitation. We also assess the accuracy of one mitochondrial (COI) and one nuclear (ITS) barcode for fast and inexpensive species delineation in the group. Our genomic data strongly support two morphologically cryptic but ecologically divergent lineages, mainly restricted to the central-eastern and western parts of the Iberian Peninsula, respectively. Larger matrices with more missing data showed increased genomic diversity, supporting that bioinformatic strategies to maximize matrix completion disproportionately exclude loci with the highest mutation rates. Moderate loci filtering gave the best results across analyses: although larger matrices returned concatenated phylogenies with higher support, middle-sized matrices performed better in genetic structure analyses. COI displayed high diversity and a conspicuous barcode gap, revealing 13 mitochondrial lineages. Mitonuclear discordance is consistent with ancestral isolation in multiple groups, probably in glacial refugia, followed by range expansion and secondary contact that produced genomic homogenization. Several apparently (unidirectionally) introgressed specimens further challenge the accuracy of species identification through mitochondrial barcodes in the group. Conversely, ITS failed to separate both lineages of Z. styliferum. This study shows an extreme case of mitonuclear discordance that highlights the limitations of single molecular barcodes for species delimitation, even in presence of distinct barcode gaps, and brings new light on the effects of parameterization on shallow-divergence studies using RAD data.