Of course even a combination of both audio and visual CAPTCHA is going to be inaccessible to a certain proportion of people, and audio CAPTCHA is equally vulnerable to this kind of exploit as the purely visual method.

IP throttling schemes will also fail in the face of a concerted "attack" from someone who avails themselves of the large number of open HTTP proxies (either mis-configured or opened up by some malware.) In checking hosts involved in reports we have seen on the NMS mailling list it could be that upwards of half of them are known open proxies or otherwise exploited hosts.

The NMS TFmail program implements a DNSBL type technique to protect from open proxies, exploited machines and other known abusive hosts: there is a bit of background in my talk from yapc::Europe last year.

To be honest you could do worse than using the TFmail rather than writing your own "contact form" program as we are actively (if fitfully) developing it and are keen to implement more "attack mitigation" measures in the future.

/J\