In the specific case of cryptography where it's unknown how bulletproof the algorithm will be why not use multiple hash functions? Perhaps using the top 10 best hash functions of the day. That way you're not putting all your eggs in one basket and if nefarious collisions are able to be created in the future you still have the other hash functions to both "trust" and double check against. It's even more unlikely that nefarious collisions will be able to be constructed that collide all the other hash functions as well. You could just append the hashes to each other or put them in a hash table or something. Maybe my computer science is not up to snuff but it seems like this would provide more resiliency against future and non-public mathematical breakthroughs as well as increased computing power such as quantum computing. Yes, it would take a little longer to compute all the hashes in day to day use, but with the benefit of a more robust system both now and in the future.