You are confusing quaternions with vectors in physical space. As the comments say, there is no such thing as having a quaternion "in" a certain frame anymore than there is such a thing as having a rotation matrix in a certain frame--both of these transform between frames and their components depend only on the relative rotation of the starting and ending frames. If $q:0\rightarrow 1$ and $r:0\rightarrow 2$ and my guess is you are looking for some third quaternion $p:1\rightarrow 2$ then since $r = pq$ you have $p = q^{-1}r = q^{*}r$ (since all rotations are unit quaternions). Note that for given $q$ and $r$ this is totally unambiguous since $q^{-1}r \neq rq^{-1}$ or any other weird permutation of $q$, $r$ and the inverse map: $q^{-1}r$ and only $q^{-1}r$ delivers the quaternion you are looking for.
Finally, to clarify some terminology, this is not the quaternion difference, which is perfectly well-defined using the Hamilton algebra (though useless for rotational analysis), but rather the relative rotation between frames 1 and 2. The same idea would hold were you to represent the rotations as matrices.