The anisotropy of the volumetric data set shouldn't be an issue, as long as your viewing volume is of cubic shape. Think of your data set as a 3D texture, which is stretched over a 3D cube with the same dimensions in all directions. You don't have to modify the 3D data slices for that, everything is done via the texture coordinates. The stretching and filtering itself is done internally, by the 3D hardware.

Of course, this implies that you actually use 3D hardware for the volume visualization, and not some other kind of software approach (eg. voxel rendering). If you don't, and you really need to interpolate the missing slices, then many interpolation techniques can be used: bilinear, bicubic, cosine, spline, etc. Each of them vary in speed vs. simplicity vs. quality metrics. But keep in mind, that you will tremendeously increase the size of your data set, if you interpolate the 'missing' parts. The best approach is to use 3D hardware, and let it to the interpolation on the fly, while rendering.