The ever-growing popularity of Wireless Local Area Networks (WLANs) in home, public, and work environments is fuelling the need for WLANs that can accommodate more stations, each with higher throughput. This typically results in WLANs containing a larger number of heterogeneous devices, making the prediction of the network's behavior and its efficient configuration an even more elaborate problem. In this paper, we propose a Markovian model that predicts the throughput achieved by each Access Point (AP) of the WLAN as a function of the network's topology and the AP's throughput demands. By means of simulation, we show that our model achieves mean relative errors of around 10% for networks of different sizes and with diverse node configurations. The model is adapted to the specification of IEEE 802.11 standards that implement channel bonding, namely 802.11n/ac/ax, and as such it can be used to provide insight into issues of channel assignment when using channel bonding. We derive guidelines on the best practice in static channel bonding given a performance metric and for different node characteristics such as the Modulation and Coding Scheme (MCS) indexes, frame aggregation rates, saturation levels, and network topologies. We then put our findings to the test by identifying the optimal channel bonding combination in an 802.11ac WLAN containing nodes with diverse characteristics. We conclude that the optimal solution is highly dependent on the particular network configuration. However, we find that, in general, larger channels are better suited for throughput maximization and smaller (and separate) channels render higher fairness.