You should contact the people who developed the software, this is a problem with that software, not with dcm2niix. This Matlab code demonstrates how to detect the true distance between voxels using the NIfTI SForm. The script below simply calculates the mm coordinates for a source voxel as well as the neighboring voxels located left, anterior and superior. It then uses Pythagorean theorem to compute the distance between the source voxel and each of its neighbors:

Input:
 mm = (hdr.mat * [0 0 0 1; 1 0 0 1; 0 1 0 1; 0 0 1 1]'); %mm position of slices
 Xmm = sqrt((mm(1,1)-mm(1,2))^2 + (mm(2,1)-mm(2,2))^2 + (mm(3,1)-mm(3,2))^2) %distance between slices
 Ymm = sqrt((mm(1,1)-mm(1,3))^2 + (mm(2,1)-mm(2,3))^2 + (mm(3,1)-mm(3,3))^2) %distance between slices
 Zmm = sqrt((mm(1,1)-mm(1,4))^2 + (mm(2,1)-mm(2,4))^2 + (mm(3,1)-mm(3,4))^2) %distance between slices

Output:
 Xmm =1.9998
 Ymm =2.0000
 Zmm =2.0000

So your voxels are still isotropic, but the image space is not aligned to world space (the MRI was acquired rotated with respect to the scanner bore).