If it isn't tight, it cannot extract a solution, of course, as there is no solution achieving that super-optimal objective value.
The moment relaxation is the dual of the sos, basically.If you look at the semidefinite program that the sos leads to, and then look its dual, you have the model you would have derived if you took the moment route. Effectively, this means that as we can extract a solution from the primal moment matrix, we could extract a solution from the sos also, as sdp solvers compute primal-dual solution. YMIP doesn't do that though. More natural/easy to do it on the moment side.