I recently briefly described the "blinding multiplication" flaw of NH-approach hash functions. This seems to be an obvious thing, but many users do not know about it and have no idea about the possible consequences.

This looks to me to be a rather serious drawback, despite the low probability, since some of the data does not affect the hash value at all (e.g. wyhash case).

I think we need to come up with some kind of test that would allow us to clearly identify such flaws.