When you think of an atlas, you can never think of it in isolation. It is always tied to a particular space, or more specifically template.
The standard Brainnetome and Glasser atlases can only be overlaid and used on the MNI template (probably specifically the MNI 2009c, as you note, because “MNI” has become a family of spaces, grids and templates).
The voxel dimensions and matrix dimensions of the Brainnetome and MNI_Glasser* atlases appear to be:
$ 3dinfo -ad3 -n4 -prefix BN_Atlas_246_1mm.nii.gz MNI_Glasser_HCP_v1.0.nii.gz
1.000000 1.000000 1.000000 182 218 182 1 BN_Atlas_246_1mm.nii.gz
1.000000 1.000000 1.000000 256 256 256 1 MNI_Glasser_HCP_v1.0.nii.gz
The Glasser atlas matches the MNI template well, but the grid dimensions are a bit different than MNI2009c. To make it match better the @SSwarper base volume MNI, I once ran:
3dZeropad -master MNI152_2009_template_SSW.nii.gz -prefix MNI_Glasser_HCP_v1.0_LPI_2009c.nii.gz MNI_Glasser_HCP_v1.0_LPI.nii.gz
3drefit -copytables MNI_Glasser_HCP_v1.0.nii.gz MNI_Glasser_HCP_v1.0_LPI_2009c.nii.gz
3drefit -cmap INT_CMAP MNI_Glasser_HCP_v1.0_LPI_2009c.nii.gz
… where the 3drefits are to copy of the labeltables and then to make the default colorbar be "integer"y, respectively.
The BN_Atlas_246_1mm.nii.gz appears to mostly match the MNI2009c well, but at the edges it seems a bit “rounded in”, oddly.
Re. Q1) Yes.
Re. Q2) It would be non-optimal re warp all functionals from TT27 space to MNI space, because they have already been re-gridded once from their native space, and warping them all again would smooth them additionally. It would be better to use the appropriate MNI template as a reference base for the dsets directly (e.g., make it the template base in @SSwarper for skullstripping and nonlinear warping and then providing that info to afni_proc.py).
Re. Q3) One would want to have a good atlas in that TT27 template space. Offhand, I don’t know of one (Daniel might?).
→ One could warp one of the MNI templates to TT27 space, and then use it there. One will have to do a bit of regridding and ROI smoothing; that might be better than warping the TT27 space EPI data to MNI, because the ROI voxels are smaller.