The estimation of robust and accurate measures of branch support has proven challenging in the era of phylogenomics. In datasets of potentially millions of sites, bootstrap support for bifurcating relationships around very short internal branches can be inappropriately inflated. Such over-estimation of branch support may be particularly problematic in rapid radiations, where phylogenetic signal is low and incomplete lineage sorting severe. Here, we explore this issue by comparing various branch support estimates under both concatenated and coalescent frameworks, in the recent radiation Australo-Papuan murine rodents (Muridae: Hydromyini). Using nucleotide sequence data from 1245 independent loci and several phylogenomic inference methods, we unequivocally resolve the majority of genus-level relationships within Hydromyini. However, at four nodes we recover inconsistency in branch support estimates both within and among concatenated and coalescent approaches. In most cases, concatenated likelihood approaches using standard fast bootstrap algorithms did not detect any uncertainty at these four nodes, regardless of partitioning strategy. However, we found this could be overcome with two-stage resampling, i.e. across genes and sites within genes (using -bsam GENESITE in IQtree). In addition, low confidence at recalcitrant nodes was recovered using UFBoot2, a recent revision to the bootstrap protocol in IQtree, but this depended on partitioning strategy. Summary coalescent approaches also failed to detect uncertainty under some circumstances. For each of four recalcitrant nodes, an equivalent (or close to equivalent) number of genes were in strong support (> 75% bootstrap) of both the primary and at least one alternative topological hypothesis, suggesting notable phylogenetic conflict among loci not detected using some standard branch support metrics. Recent debate has focused on the appropriateness of concatenated versus multi-genealogical approaches to resolving species relationships, but less so on accurately estimating uncertainty in large datasets. Our results demonstrate the importance of employing multiple approaches when assessing confidence, and highlight the need for greater attention to the development of robust measures of uncertainty in the era of phylogenomics.