The 3dvolreg help states 40 bytes will be used for each voxel in one of the input datasets. With your data resampled to 0.5mm^3, the size of the data is about 512 MB or 256 million voxels. That's about 10GB required to do the volume registration. Unfortunately, 3dvolreg, 3dWarpDrive and 3dAllineate all fail with that size dataset. You can (and probably should) reduce the size by setting a larger voxel size in the 3dWarp step to a larger size. Resampling to 1mm^3 works on my system, and is probably adequate for you.