Efficient and accurate radio propagation modeling is essential for optimization of both radio sensing and communication systems. However, highly accurate full-wave methods remain inefficient at high frequencies, as unit of computation has to be made much smaller than the wavelength. On the other hand, ray-based approaches offer the desired speed, but the surface element (typically, a triangle) must be made much larger than the wavelength, thus making it difficult to represent complex curved surfaces of common objects, such as cars or unmanned aerial vehicles. As a result, for super high frequency (SHF)/extremely high frequency (EHF) bands, it is challenging to select a method that is both fast and capable of capturing curved surfaces correctly. To address this issue, we present a method that offers a reasonable tradeoff between speed and accuracy for radio propagation modeling in the bands of interest. Specifically, we combine efficient voxel scene representation targeting a cone tracing algorithm with a statistical scattering model. To confirm the validity of our approach, we report the dependence of reflected power on the distance for basic primitives, such as cone and sphere, for which closed-form radar cross-section solutions are known. We demonstrate that the proposed approach is superior compared to the conventional ray-tracing solutions, while posing no restrictions for efficient voxel cone tracing implementations.