The ability to load datasets in spatial fractions would be somewhat tricky to implement, given the structure of AFNI dataset I/O (which is 3D volume oriented). Another big memory consumer in 3dDeconvolve is space for all the output volumes -- those would have to be shuttled off to temporary disk files, as well, adding to the complexity of programming required. These reasons are why I've not wanted to tackle this issue, preferring to outsource my suffering to the users via 3dZcutup and 3dZcat.