I'm of the same mindset. I mean -- surely there are some collisions always. And I actually liked the property of the original version that a zero value hashed t

I'm of the same mindset. I mean -- surely there are some collisions always. And I actually liked the property of the original version that a zero value hashed to all zeroes. While the rest of the pre-image was shuffled. But I suppose that being able to compute collisions is not a great thing. You may think that if it is not intended to be used in a cryptographic sense it does not matter. However, if you can easily compute collisions you could, theoretically of course, launch a denial-of-service attacks on a service (cache denial) by sending lots of values to the same hash slot. I mean, with probing and so on there are ways around this and in the end maybe the conventional wisdom ought to be revised, maybe there isn't so much to worry about from a few collisions, even easy to compute ones. But, as it stands, right now, the market demands collision resistance. So I damn well gave it to them.