I've just run some more tests with hashing and things don't seem to be so good... That code I was so proud of when I was writing this question is running only 1.5 times faster than the old one. The only difference are calls to OpenSSL's MD5 function and to memcmp. All in all, no x7 increase... What a pity...