OK, here's what I currently use for that:
3dWarpDrive warps the EPI dataset spatially, and can save the rotation parameters in a 1D file.
I use the rotation angles in that 1D file to be entered in my transformation equation (see basic math how to rotate vectors, you'll have a bunch of sin and cos functios).
With 3dcalc I then calculate the rotated tensor gradient basis (note 3dcalc doesn't like only 1D files as input, so I faked it by using a small dummy 3D brik as the first input, which will never be used in the calculation).
Using this rotated gradient file in 3dDWItoDT gets me the appropriate tensor directions in the result.
Greetings from HBM,
Wolfgang