Any "higher order" interpolation or filtering scheme (which includes the methods used in 3dFourier) has the potential for producing values that exceed the min..max range of the data. What is probably happening in your data is that values near the edge of the brain have very low intensity (min near 0), so going below that range is obvious. In the center of the brain, there are undoubtedly values below the local min value, but those aren't so obvious (372 doesn't look so different than 400, for example).
If you wish to clip off these negative annoyances, you can do so with a 3dcalc command of the form
3dcalc -a zork+orig -expr 'step(a)*a' -prefix zork_clipped
bob cox