As far as I know outputs in T1w space should have the same resolution (voxel size) as outputs in the MNI space. If they don’t (please check) that’s a bug that needs to be fixed. For reference this is where in code we use custom reference image to get maching voxel size: https://github.com/poldracklab/fmriprep/blob/master/fmriprep/workflows/bold.py#L899
If this is not a bug and indeed voxel sized in T1w and MNI outputs are the same there might be another explanation for the file size difference. The field of view of your T1w image might be larger than the field of view of the MNI template (thus leading to more voxels and larger size). Additionally it could be the case that due to yet another bug MNI outputs are in float64 and T1w outputs are in float32.
So help us debug this it would be great if you could share the following info for both MNI and T1w outputs: resolution (voxel size), field of view (number of voxels in each dimension) and data type. It’s easy to get this data using fslhd.