The Sediment Delivery Ratio (SDR) is ratio of net to gross erosion (or actual to theoretical erosion), where to my way of thinking, gross erosion is the erosion that takes place at a point on the landscape, and net erosion is the sediment that actually makes it out of an HRU, subbasin, or watershed (choose your scale) after some of the particles have been trapped along the flow path. The longer the flow path, the more sediment traps are likely to be encountered, and hence net erosion (per unit area) and SDR get smaller as the area increases (from HRU -> subbasin -> watershed; there are some exceptions). The MUSLE equation gives what it thinks is the actual erosion leaving an HRU, prior to any wetland sediment traps or channel processes (including reservoir sedimentation).
The USLE t/ha value given in the output is not used in SWAT calculations but is given for comparative purposes. It tends to be larger than the MUSLE t/ha. I presume this is because the USLE was (I think) derived from plot-scale studies, and the MUSLE was derived with data from actual small watersheds, and so the MUSLE already takes into account some sediment trapping at the small watershed (subbasin?) scale, or at least at the field (HRU?) scale.
I suppose you could calculate SDR as MUSLE t/ha divided by USLE t/ha, which would give you an SDR at the HRU scale (you could expand this to the subbasin and watershed scales as well). I presume others have done this, though I haven't done a Google Scholar search on the subject. If you're going to do that calculation, then you should do the literature search to see what others have done.
Cheers,
-- Jim