Just to add a little more background to this,
the +tlrc.HEAD file does not contain the pieces
of the 2 transformations, it contains the complete
ORIG->TLRC transformation.
It would be possible to have adwarp figure out
that the acpc dataset is half way (e.g. by noting
the spaces and that they share a warp parent),
and then adwarp would have to compute the
tlrc warp of (the inverse acpc warp of (the acpc data)).
Again, there are multiple ways that the above
could be applied (besides in adwarp, which might
lead to actual effort, which is obviously bad).
So depending on how you originally populated
those datasets, perhaps there are easier routes.
- rick