Well they have put developer time in, not to mention massive amounts of computing power. It processes every video uploaded to youtube, and compares it to every video in the contentID system. It's pretty amazing. There's a short TED talk on it http://www.ted.com/talks/margaret_stewart_how_youtube_thinks...
I'd argue that with the level of false positives it needs improvement.
I won't argue that the existing system isn't both impressive in terms of scale and complexity, just that it's not working as well as it should over something like blocking ma & pa's home video for the Britney Spears track you can hear in the background.
