My software will losslessly rotate 3D acquisitions to the nearest orthogonal angle and save this as a 'o', the change is also reflected in the NIfTI header SForm and QForm. The raw and orthogonal images will look identical with tools that reorient data based on the transform (e.g. MRIcroGL and SPM), however the 'o' images tend to look more correct in tools that ignore these transforms (e.g. MIPAV and FSLview).

By the way, whenever you run dcm2nii it will advise you to upgrade to dcm2niix (which comes with MRIcroGL). The vendors are constantly reinterpreting the DICOM standard, and if you use old tools to convert new datasets it may not function as you expect. While I hope my old tools are mature and stable, I have moved my development efforts to newer tools. There was no way I could have predicted new changes in DICOM usage when I wrote that code.