I am not sure what you are trying to do with the motion correction. It looks like you are trying to run the equivalent of 3dvolreg twice on the data. If that's the case then there is nothing to gain there and you should not be doing it.

If you have a lot of motion, as indicated by 3dvolreg's output, and you want to discard such points, do so with the censor option in 3dDeconvolve. I recommend that you get a visual confirmation of large motion artifacts by viewing the time series dataset in video mode.
Launch AFNI
Switch Underlay to Time Series
Open an image window or two
Open a graph window
With the mouse pointer on the graph window, hit 'v' on the keyboard
watch a video of the images over time
to stop the video, press 'space bar' with the mouse pointer on the graph window


cheers,
-G