I do not see why this would be needed. I think resampling to a uniform matrix is important while extracting texture features, but here only shape and firstorder are extracted.
This 3x3x3 resampling is only done if you use the --fast option. This greatly increases runtime and the resulting volume. Otherwise the original resolution is used.
It appears that the image is resampled to 3x3x3 prior to the extraction of the radiomics features:
https://github.com/wasserth/TotalSegmentator/blob/master/totalsegmentator/statistics.py#L29
I do not see why this would be needed. I think resampling to a uniform matrix is important while extracting texture features, but here only shape and firstorder are extracted.