It can certainly happen for a variety of reasons. Some potential questions:
1) What's different between your acquisition and that of your colleague?
2) Have you checked the alignment of your data to the anatomical?
3) Does your data have the slice timing built in correctly? Can you post the output log?
I'd recommend adding some things to your proc script to possibly account for motion or registration:
-regress_censor_motion 0.2
-regress_censor_outliers 0.1
-tlrc_NL_warp
Things you could remove:
-align_opts_aea -deoblique on
