This paper provides several optimizations of the rank-based approach for complementing B{u}chi automata. We start with Schewes theoretically optimal construction and develop a set of techniques for pruning its state space that are key to obtaining small complement automata in practice. In particular, the reductions (except one) have the property that they preserve (at least some) so-called super-tight runs, which are runs whose ranking is as tight as possible. Our evaluation on a large benchmark shows that the optimizations indeed significantly help the rank-based approach and that, in a large number of cases, the obtained complement is the smallest from those produced by a large number of state-of-the-art tools for B{u}chi complementation.