Hi Michael

If the data is still showing some consistent biases, that indicates that there are some remaining global effects which have not been sufficiently compensated for, so yes, I would recommend applying additional denoising procedures until the data looks artifact-free (that includes for example outlier detection/scrubbing, additional aCompCor dimensions, additional first-order temporal derivatives and/or second-order powers of the movement parameters, and/or ICA-based denoising components). In any way, the observation that the biases look significantly different in the spatially-unsmoothed vs. spatially-smoothed data also sounds somewhat unexpected, so it may be worth double-checking to make sure that all of the data has been properly processed. Also, the observation that an additional aCompCor procedure still results in a significant effect on the histograms after an initial aCompCor step had been run on the original data looks really unexpected as well, so it might also be worth trying to identify why this is the case (e.g. are you sure that the proper/same white/CSF masks had been used in both cases)

Also a general comment/recommendation, as with any other dataset, focus first on trying to clean up your data and identifying the source of potential artifacts before looking at any of your planned analyses or results (e.g. define the choice of denoising steps that seems to result in "best" removal of biases and minimal differences between-subjects in the voxel-to-voxel histograms, and once you are satisfied with this move on to your planned first-level analyses).

Hope this helps
Alfonso