Well, it still wouldn't be exactly the same, though. As I understand it, the default weighting is a "smoothed" version of the data. If I use a masked version of the data, the smoothing will not be performed. May not matter, but...
To follow up on the core dump - it might also be worth mentioning that the mask file I was using is all zeros for most slices, with a couple of masked areas on one slice. Not sure if that matters.