While recent literature has clearly demonstrated that an extensive personalization of the musculoskeletal models was necessary to reach high accuracy, several components of the generic models may be further investigated before defining subject-specific parameters. Among others, the choice in muscular geometry and thus the level of muscular redundancy in the model may have a noticeable influence on the predicted musculotendon and joint contact forces. In this context, the aim of this study was to investigate if the level of muscular redundancy can contribute or not to reduce inaccuracies in tibiofemoral contact forces predictions. For that, the dataset disseminated through the Sixth Grand Challenge Competition to Predict In Vivo Knee Loads was applied to a versatile 3D lower limb musculoskeletal model in which two muscular geometries (i.e., two different levels of muscular redundancy) were implemented. This dataset provides tibiofemoral implant measurements for both medial and lateral compartments and thus allows evaluation of the validity of the model predictions. The results suggest that an increase of the level of muscular redundancy corresponds to a better accuracy of total tibiofemoral contact force whatever the gait pattern investigated. However, the medial and lateral contact forces ratio and accuracy were not necessarily improved when increasing the level of muscular redundancy and may thus be attributed to other parameters such as the location of contact points. To conclude, the muscular geometry, among other components of the generic model, has a noticeable impact on joint contact forces predictions and may thus be correctly chosen even before trying to personalize the model.