It has recently been shown that cut-free deep inference systems exhibit an exponential speed-up over cut-free sequent systems, in terms of proof size. While this is good for proof complexity, there remains the problem of typically high proof search non-determinism induced by the deep inference methodology: the higher the depth of inference, the higher the non-determinism. In this work we improve on the proof search side by demonstrating that, for propositional logic, the same exponential speed-up in proof size can be obtained in bounded-depth cut-free systems. These systems retain the top-down symmetry of deep inference, but can otherwise be designed at the same depth level of sequent systems. As a result the non-determinism arising from the choice of rules at each stage of a proof is smaller than that of unbounded deep inference, while still giving access to the short proofs of deep inference.