Alright,

I have figured out the problem.  

I looked at sfnwmrda1779922_session_1_rest_1.nii.gz and its corresponding AAL timecourses.  The first of which is for ROI #2001. In the attached plot, the black line corresponds to the mean timecourse for this ROI extracted from snwmrda1779922_session_1_rest_1.nii.gz, which is correctly detrended.  The red line corresponds to the averaged band passed filtered time course from sfnwmrda1779922_session_1_rest_1_AAL_TCs.1D.  The blue line is the correctly bandpassed filtered time course.

The problem is with how I am calling 3dFourier to do the bandpass filtering, I used the -retrend option, which is described as:
"Any mean and linear trend are removed before filtering. This will restore the trend after filtering.".  I used this option as a carryover from the days when I bandpassed filtered (haven't used that in a while) and also wanted to preserve the image detail.  This is what is causing the odd behavior.

The odd thing is that the file that is the input to 3dFourier is nwmrda1779922_session_1_rest_1.nii.gz, which has already been detrended.  I verified this by re-detrending the black line in the plot, and got the black line again.  So somehow 3dFourier is finding a bogus trend in the data.  My guess is that what it is doing is fitting a straight line through the first point, which is -1 and the last point, which is 2 and ignoring everything in between.  

Interestingly, there is a new discussion on the AFNI forum about 3dFourier weirdness.

Thank you very much Cheng for finding this problem.  I will fix the data and put it on NITRC soon.

Cheers,
Cameron