In the microcirculation, a plasma layer forms near the vessel walls that is free of red blood cells (RBCs). This region, often termed as the cell-free layer (CFL), plays important haemorheological and biophysical roles, and has been the subject of extensive research. Many previous studies have considered the CFL development in single, isolated vessels that are straight tubes or channels, as well as in isolated bifurcations and mergers. In the body, blood vessels are typically winding and sequentially bifurcate into smaller vessels or merge to form larger vessels. Because of this geometric complexity, the CFL in vivo is three-dimensional (3D) and asymmetric, unlike in fully developed flow in straight tubes. The three-dimensionality of the CFL as it develops in a vascular network, and the underlying hydrodynamic mechanisms, are not well understood. Using a high-fidelity model of cellular-scale blood flow in microvascular networks with in vivo-like topologies, we present a detailed analysis of the fully 3D and asymmetric nature of the CFL in such networks. We show that the CFL significantly varies over different aspects of the networks. Along the vessel lengths, such variations are predominantly non-monotonic, which indicates that the CFL profiles do not simply become more symmetric over the length as they would in straight vessels. We show that vessel tortuosity causes the CFL to become more asymmetric along the length. We specifically identify a curvature-induced migration of the RBCs as the underlying mechanism of increased asymmetry in curved vessels. The vascular bifurcations and mergers are also seen to change the CFL profiles, and in the majority of them the CFL becomes more asymmetric. For most bifurcations, this is generally observed to occur such that the CFL downstream narrows on the side of the vessel nearest the upstream bifurcation, and widens on the other side. The 3D aspects of such behaviour are elucidated. For many bifurcations, a discrepancy exists between the CFL in the daughter vessels, which arises from a disproportionate partitioning between the flow rate and RBC flux. For most mergers, the downstream CFL narrows in the plane of the merger, but widens away from this plane. The dominant mechanism by which such changes occur is identified as the geometric focusing of the two merging streams. To our knowledge, this work provides the first simulation-based analysis of the 3D CFL structure in complex in vivo-like microvascular networks, including the hydrodynamic origins of the observed behaviour.