Species tree estimation is a basic part of much biological research, but can be challenging because of gene duplication and loss (GDL), which results in genes that can appear more than once in a given genome. To construct species trees from gene families, biologists typically pick one copy of each gene in each species (based on estimations of orthology), but orthology detection is not yet reliably accurate, and incorrect orthology inferences can result in incorrect species trees. Restricting datasets to single-copy genes is another common solution, but this reduces the amount of available data, which is undesirable (and even genes that appear to be single-copy may have evolved with duplication and loss). Thus, all common approaches in phylogenomics reduce available data and are error-prone, and methods that do not discard data and have high accuracy on large heterogeneous datasets are needed. Here, we present FastMulRFS, a method for estimating species trees from multrees. We prove that FastMulRFS is polynomial time and statistically consistent under a generic model of gene duplication and loss provided that only duplications occur or only losses occur. We present the results of an extensive simulation study where gene tree heterogeneity is due to gene duplication and loss, incomplete lineage sorting,and gene tree estimation error, and show that FastMulRFS matches the accuracy of MulRF (which tries to solve the same optimization problem) and has better accuracy than ASTRAL-multi (which is statistically consistent under GDL), while being much faster than both methods.