Purpose: In a previous work, a PENELOPE Monte-Carlo (MC) model of a Cyberknife system equipped with fixed collimator was developed and validated for in-field dose evaluation. This model was then extended to precisely assess out-of-field doses and was validated against 1-D measurements. The aims of this work are firstly to validate our model in 3-D and in heterogeneities and secondly to evaluate the precision of the treatment planning system (TPS) Multiplan for off-axis doses. Methods: A treatment plan, mainly using 40 mm collimator, has been simulated for a heterogeneous configuration of the Easycube phantom using the Multiplan TPS. Each generated treatment beam was also simulated with our extended Cyberknife Monte-Carlo model. Nine EBT3 films, calibrated using a rigorous protocol, were placed between the phantom slices and irradiated according to the plan. Dosimetric data such as DVH or planar doses obtained using our MC model were compared with measurements for validation and with TPS doses (calculated with the TPS MC algorithm and RayTracing) to assess their accuracy. Results: A good agreement was obtained between film measurements and our MC simulation. However, the TPS MC algorithm underestimates doses outside the beams up to 45% at 6 cm from the PTV edge and up to 18% for the mean dose of a structure placed outside the beams near the PTV. RayTracing highly overestimates doses far from the PTV (up to 310% at the furthest film level) as it does not model particle transport. Conclusion: Our MC model was successfully validated in 3-D. The TPS algorithms do not correctly model doses delivered outside the beams, even close to the PTV, and for large field sizes (for which phantom scatter is predominant). A study using smaller field sizes is ongoing to investigate the limits of the TPS when determining out-of-field doses.