The relevancy algorithm also solves it partly when you upvote and downvote enough things and the system build can build up some profile vectors for you to improve ranking for yourself.
We did some really through analysis and built a bunch of statistical models and tests. We think we can handle the problem makers and still try to maintain highly personal relevant and realtime posts wherever you are at.
As a test, we scrapped 4chan and dumped posts of that scum at random intervals. Beta users killed it really fast and quality didn't suffer. Turned it into a game a little bit and it reveled our early evangelists.
The trick is doing all this relevancy matching without crippling scaling issues when we turn up the dial on users. Really fun engineering problem.
This is not a unique criticism - all social media is subject to attack cases like this - the difference here would be that pictures tend to carry more weight than a simple textual message.