You have a mask in +tlrc space and time series in +orig space, so you have to convert something. You can do it either way, as you guess:
mask -> +orig using 3dfractionize -warp
time series -> +tlrc using adwarp (or the AFNI GUI's "Write")
followed by 3dmaskave, 3dmaskdump, and/or 3dROIstats, as appropriate.
I'm not sure that it makes much difference which way you go. You might try both on one dataset to see if you get significantly different answers (I'm assuming you want to create an ROI-averaged time series or some similar statistic).
A practical point is that each 3D+time dataset in +tlrc space will be huge (100s of MB), so that the second method will be slower and consume more disk space.
bob cox