From all the comments and what I have noticed it sounds like a good question is "Is it better to have false positives or false negatives" in the spam problem. I personally think that its better to have false negatives then positives and a lot of the comments here seem to reflect that.
That depends on how many false positives and negatives we have -- you don't really know how many of either. There are 100 million hosts in our crawl, how can you estimate if we have a false positive problem from looking at the list of the top 100 marked spam?
Your right. Would there be a way to get a sample of the database and attempt to do a test if there are a considerable amount of false positives?