I have modified 3dAutomask slightly to allow you to specify the 'clip level fraction' used to set the initial clipping level for the masking. A smaller fraction means that more voxels will be included in the mask. The default value, formerly hardwired into the program, was 0.5 (cf. program 3dClipLevel for details on the clipping algorithm).
Attached is an image of the automask with -clfrac 0.5 (the default) and -clfrac 0.25 -- voxels in red are in both masks, and those in yellow in just the -clfrac 0.25 mask. You can see how the smaller clip fraction expands the mask, but not uniformly like -dilate does.
This change is in the AFNI source code now. It will be compiled tonight into the binary packages. See if this mollifies you.
However, using -dilate as you are doing won't really have a lot of impact on the normalization, since that is done on a voxel-by-voxel basis.