On normalization of proofs in set theory