At present, AFNI does not directly support any statistic whose parameters vary from voxel to voxel. This is the main reason why the unpooled variance t test is not advised, since there is no easy way to threshold it coherently. It would be possible to modify 3dttest to save the DOF parameter per voxel, but then it would be necessary to use 3dcalc to convert (t,DOF) pair to some other statistic (e.g., a z score) for coherent thresholding and inference.
bob cox