The voxel-based indirect boundary element method (IBEM) combined with the Laplace-kernel fast multipole method (FMM) is capable of analyzing relatively large-scale problems. A typical application of the IBEM is the electric field analysis in virtual-human models such as the model called Duke provided by the foundation for research on information technologies in society (IT'IS Foundation). An important property of voxel-version Duke models is that they have various voxel sizes but the same structural feature. This property is useful for examining the O(N) and O(D-2) dependencies of the calculation times and the amount of memory required by the FMM-IBEM, where N and D are the number of boundary elements and the reciprocal of the voxel-side length, respectively. In this paper, the dependencies were confirmed by analyzing Duke models with voxel-side lengths of 5.0, 2.0, LO, and 0.5 mm. The finest model had 2.2 billion voxels and 61 million square elements. In addition, a technique that improves the convergence performance of the linear equation solver by considering the non-uniqueness of the electric potential is proposed, and its effectiveness is demonstrated.