My initial thoughts are that it does matter. From the theory, what we observe at a pixel is the aggregation of responses from the surface coverage of the pixel. Since this aggregation is nonlinear, we may be unable to accurately interpolate fine resolution information from coarser resolution. However, we can aggregate finer resolution information to obtain a fair representation of the information at the coarser resolution. I know experts in mixed pixel research may not agree with me entirely but from my experience, it is easier to move from fine to coarse than from coarse to fine.