There's another more practical reason it's hard to kill mean comments: they often have replies by the time I see them. It would mess up the conversation if they got taken out then.
Once an individual has a given number of penalty boxes, a moderator considers whether to ban them. The danger of doing stuff like this is that subjects that interest this community inevitably also spill over into the political realm and as others have pointed out the use of up votes and downvotes are ambiguous and further, open to abuse.
If you want to publicly shame someone, then once a member is on probation after a certain number of penalties/flags, add a dot or something relatively benign in color to their name.
Edit - I'd add that if it were up to me, I'd try to control personal attacks above all else given the community is pretty good at calling people out on comments that add little to no value.
With a huge crowd, the system needs to train the users by providing cues and publicly "punishing" (through greyed-out text and negative scores) ill behavior. The more people you have, the more important this feedback system becomes.
Then you need to bring onto your team some group of curators/moderators whose number 1 task is to keep the site free of nastiness on a nonpartisan, equal opportunity basis.
The only way to keep human moderation workable is to keep the work of each individual moderator small and scale out tree-style. Perhaps that will be the ultimate shape of the Web: a set of top level news sites, aggregating the best commentary from a larger set of second-tier news sites, aggregating the best commentary from a larger set of tertiary news sites...
Lots of communities have tried to push moderation into the hands of users via flagging, and it never seems to work. I think the best and quickest solution to the problem is to use trusted but invisible moderators.