Hello Ping,
You could merge the two datasets but given that 3dMVM is univariate (voxelwise), you won't get different results by combining the two hemispheres, except for FDR corrections. Nonetheless, you can still combine the two with something of this sort:
1- use FreeSurfer's mris_convert to write out the datasets in gifti format (.gii)
2- use ConvertDset to write out the two volumes in .1D format - make sure you are not dealing with sparse datasets and if you are, tell ConvertDset to pad the output appropriately for each of the hemispheres: see -pad_to_node option
3- Vertically catenate the two .1D datasets (say L.1D.dset), and R.1D.dset) nto 1 LR dataset with (tcsh syntax):
1dcat L.1D.dset > LR.1D.dset
1dcat R.1D.dset >> LR.1D.dset
4- Create an LR surface just to be sure you are not making an atrocious mistake. Here I assume you have run on FreeSurfer's fsaverage surfaces: @SUMA_Make_Spec_FS -GIFTI -sid fsaverage6
cd ....../SUMA/
ConvertSurface -i lh.smoothwm.gii -i rh.smoothwm.gii -merge_surfs -o_gii lrh.smoothwm.gii
5- Make sure that when you load LR.1D.dset onto lrh.smoothwm.gii, you get what you expect. I assume you know how to do this with SUMA, otherwise let me know and I'll give you some pointers.
6- Now feed the LR.1D.dset into 3dMVM and have at it.
Note, there might be an easier way to do this with Nick's Surfing toolbox: (http://surfing.sourceforge.net/Welcome.html) but I leave it up to him to elaborate.
cheers,
Ziad