Gene duplication plays a crucial role in species adaptation and the emergence of functions, making the inference of past duplications key to understanding evolution. Whole-genome duplications (WGD), which copy all the gene families in a single event, had a significant impact on the evolution of plants, yeast, and even vertebrates. Genome-scale data is often used to infer WGDs, for instance syntenic blocks and gene counts, but ancient WGDs are notoriously difficult to retrace, as their signals get lost in rearrangements and losses. Reconciliation of species and gene phylogenies may find ancient duplications, although current tools often assume independence between gene families and can miss WGDs, in which all genes are interdependent. Phylogenomics reconciliation addresses this by reconciling multiple gene families, but current models restrict the space of possible reconciliations, ignore gene losses due to fractionation, or rely on conserved syntenies across multiple species, limiting the number of genes that can be considered simultaneously. In this work, we focus on a reconciliation model that is synteny-free, that considers losses, and that allows flexible remapping of gene duplications. Reconciliation under this model was shown NP-hard and current algorithms are far from achieving the scalability needed for phylogenomics analysis. We introduce novel algorithmic techniques that can handle large datasets containing tens of thousands of gene trees, which was not possible before. Our experiments on simulations and real data demonstrate that the traditional lca-mapping can make incorrect predictions in cases of WGD followed by fractionation, to which our approach is more robust. Tests on real data also show that our approach can recover WGDs that can be missed by other phylogenomics reconciliation methods.