yes...compressed avi questions and blendshapes....repeating like a bad a curry.
OK, the hierarchy thing....yes blendshapes work within local space....but, the problem is if they are in same hierarchy as the target, then people tend to bind the group to the skeleton...and then you get double defs. Nonquad is correctto suggest that its a precautionary measure more than anything else. Plus, you should delete history on the belndshapes.
As for what our friend here has done....well....the pivots should be in the same local space to each blendshape as the target shape...in other words DON'T MOVE THE PIVOT from where it was when you duplicate the target shape for each belndshape. Likwise, don't freeze transformations. The reason why this screws it up is that, basically, the cv's of the shape are relative to the original. So, if the original target shape has a pivot of an effective <3, 4, 2> vector from the centre of the shape within its own local space, then if you move the pivot, the cv's are in a different relative position to the target cv's. So it moves the same distance, but minus/plus the difference of pivot vector. Likewise, if you freeze the transforms, you have changed your pivot to an effective vector of <0,0,0> ....same result.
Moral of the story is, DON'T centre pivots or freeze transforms.
:>