I don't think all instrumentally equivalent theories are epistemically equivalent. There are other considerations, like simplicity, that play a role in determining which theory we should regard as true. For instance, we could posit that the geometry of our universe is actually Euclidean, and apparent non-Euclidean measurements are accounted for by the fact that the size of our measuring sticks changes depending on where they are in space. This theory is instrumentally equivalent to the theory that our universe is non-Euclidean, but the choice between them is not merely conventional. The Euclidean account is less simple, and so inferior.