It is possible that I will make the move to full NIfTI-ization in AFNI. However, before I can even consider making this momentous conversion, I need to do several things, including

(a) finalize the structure and code for storing AFNI-specific information into a NIfTI-1.1 header extension (this step is easy, and half done already);

(b) make sure that the NIfTI-1.1 reading code works well in AFNI, especially when the datasets are oblique and/or have non-orthogonal axes (this step is hard, since it requires thinking about coordinate systems, which makes my brain hurt -- I am not a natural born geometer);

(c) consider how to support (if at all) the more complex NIfTI-1.1 datasets, such as vector-valued time series, matrix-valued datasets, datasets with statistics whose parameters (e.g., DOF) varies with voxel, and so forth -- all things that AFNI doesn't allow at present.

Note that you can use almost any AFNI '3d' program now to write out a NIfTI-1.1 dataset (including to3d), if you just end the new dataset's prefix with '.nii' or '.nii.gz'. However, since step (a) is not yet finished, some specific AFNI attributes stored in a .HEAD file will not survive this process (e.g., 3dvolreg parameters, history note, Talairach markers, ID codes).