The radiation therapy specific Voxel Monte Carlo (VMC++) dose calculation algorithm achieves a dramatic improvement in MC dose calculation efficiency for radiation therapy treatment planning dose evaluation compared with other MC algorithms. This work aims to validate VMC++ for radiation therapy photon beam planning. VMC++ was validated with respect to the well-benchmarked EGS-based DOSXYZnrc by comparing depth dose and lateral profiles for field sizes ranging from for 6 and beams in a homogeneous water phantom and in a simulated bone-lung-bone phantom. Patient treatment plan dose distributions were compared for five prostate plans and five head-and-neck (H/N) plans, all using intensity-modulated radiotherapy beams. For all tests, the same incident particles were used in both codes to isolate differences due to modeling of the radiation source. Voxel-by-voxel observed differences were analyzed to distinguish between systematic and purely statistical differences. Dose-volume-histogram-derived dose indices were compared for the patient plans. For the homogeneous water phantom and the bone-lung-bone phantom, the depth dose curve predicted by VMC++ agreed with that predicted by DOSXYZnrc within expected statistical uncertainty in all voxels except the surface voxel of the water phantom, where VMC++ predicted a lower dose. When the electron cutoff parameter was decreased for both codes, the surface voxel agreed within expected statistical uncertainty. For prostate plans, the most severe difference between the codes resulted in 55% of the voxels showing a systematic difference of 0.32% of maximum dose. For H/N plans, the largest difference observed resulted in 2% of the voxels showing a systematic difference of 0.98% of maximum dose. For the prostate plans, the most severe difference in the planning target volume was 0.4%, the rectum was 0.2%, the rectum was 0.2%, the bladder was 0.3% and the bladder was 0.3%. For the H/N plans, the most severe difference in the gross tumor volume was 0.4%, the clinical target volume was 0.2%, the nodes was 0.2%, the parotids was 0.8%, and the cord was 0.8%. All of these differences are clinically insignificant. VMC++ showed an average efficiency gain over DOSXYZnrc of at least an order of magnitude without introducing significant systematic bias. VMC++ can be used for photon beam MC patient dose computations without a clinically significant loss in accuracy.