Hi there,
Yes that does seem strange now you mention it!
I double-checked my script and I am definitely using each subject's brain mask and sqrt_ResMS image to estimate their smoothing amount. So perhaps the -2difMAD option isn't appropriate or needed with this type of input?
It may actually be the case that my original estimates of 13-14mm are the correct ones - my preprocessing pipeline in SPM interpolates the images twice (once for motion and distortion correction, and once when warping to standard space).
Thanks for your help!