The increasing uses of voxelied human body geometry have created challenges in many aspects of Monte Carlo simulations. This paper presents an investigation on the maximum number of voxels the MCNP code can handle by using an extremely detailed VIP-Man model. The VIP-Man tomographic model is a radiation dosimetry model developed from the segmented Visible Human color cross sectional images. Several variables that MCNP uses to store the information related to the MCNP "universe" have been changed in this study to maximize the efficient usage of the system memory. Two modifications have been made in the MCNP source code for the variable, laf, which greatly reduce the amount of system memory required for these kinds of MCNP geometries. Another variable, mazu, which is not necessarily used when running the voxel models, is only allocated if it is needed. Without the modifications, the current release of MCNP5 crashed when lattice geometries used more than 25 million voxels. With the modifications, however, a simple voxel model was tested and results showed that the modified version of MCNP5 can handle lattice geometries of 400 million voxels in size. This modified version was used to calculate the transmission photon radiograph of a portion of the VIP-Man with 100 million voxels. From this study, we conclude that, although there have been many improvements in computer speed and tallies in the MCNP, both the current and modified versions of the MCNP remains unable to handle the whole VIP-Man model at the original voxel size of 0.33mm × 0.33 mm × 1.0 mm, with 3.7 billion voxels, which would exceed the operating system limits of 32 bit MS Windows PCs if every voxel's material was stored in memory. However, a VIP-Man with larger resolution, such as 1 mm cubic voxels, would potentially be able to run on the modified version of MCNP5.