As argued in our paper "Is SHA-1 conceptually sound?", the main problem with SHA-1 is its bad message expansion code. In "Provably Good Codes for Hash Function Design", we give a general methodology for designing extremely efficient to code quasi-cyclic codes, which have provably good minimum distance. One such code is at most 5% slower in software than the SHA-1 code, and we use it to propose a new hash function SHA1-IME (SHA-1 with improved message expansion). We claim that the new hash function is resistant to existing (including recent developments and their natural extensions) differential collision attacks. Further, as outlined in the paper "Is SHA-1 conceptually sound?", it most likely withstands any attack.
Uri Blumenthal (Intel) has written an internet draft (with test vectors) for SHA1-IME.