The question I care is: Is the probability of naturally occurring a collision in MD5 hashing significantly more than any other 128 bit hashing algorithm?
Sure, but if you're willing to accept lots collisions on "bad" input there are faster hashes (Bernstein and Jenkins have nice fast non-cryptographic hashes, for instance.)
Non-cryptographic hash functions may be a bad idea: e.g. don't store URI parameters (?foo=bar&baz=bar) in such a hash table, or you'll be vulnerable to rather simple DoS (this was all over the internet a week or two ago.)
> Non-cryptographic hash functions may be a bad idea: e.g. don't store URI parameters (?foo=bar&baz=bar) in such a hash table, or you'll be vulnerable to rather simple DoS (this was all over the internet a week or two ago.)
I think universal hashing is the usual protection against that kind of attack, and I think universal hashing is not considered cryptographic:
The answer is: No.