Hi Kevin,
Autorange value comes from the maximum value in the data set. For voxels with very little or no signal in them, the Hilbert Transform of the cross-correlation curve might not cross zero at a lag < stimulus period. Also, the algorithm stops searching for a zero crossing whenever the lag exceeds 1/3 the duration of the time course. In both cases, you'll end up with a large (but meaningless delay value which when multiplied by 2pi/period results in delays > 2 pi radians. These values are meaningless but you don't (should not) use them anyway because the cross correlation coefficient for such voxels in ~= 0 (i.e. no significant activation at that location).
You should set the range manually to 2pi for your data.
cheers.
ziad