Hi Tracy,
As Bob said, the mask should not impact the scaling (normalization) in brain regions. If your mask is too generous, you might have some freakishly large %signal changes outside the brain where the mean signal is close to 0. The safeguard I mentioned in class was to add *step(a-threshold) to your scaling equation, where a is the variable for the voxel's mean and threshold is a cut off you have to set based on the range of values outside the brain. Obviously, threshold should be quite low, something like 0.01 or less.
cheers,
-ziad