I have been interested in this question for quite some time. I focus mainly on college football and basketball. I have spent many hours analyzing a prediction tracking website, newsletters, consensus polling questions here (thanks Betallsports!), and selected posters here and tried to see whether the knowledgeable of crowds and 'experts' provides any edge. My thought going in to this was that the taking the consensus from these three independent data streams would provide additional useful information to help me with handicapping. Taking the straight results from my limited studies, I have found that there is little advantage to any one of these streams in NCAAF (~51.8%). When combined, the success rate rose slightly, but not enough to make a difference (~53.2%). NCAABB is a bit better 52.7% and 55.4%.
If I was to venture a guess, I would say that most of the problems with prediction arise because the algorithms and opinions are largely mathematical/statistical models that rely upon past performances to predict future events. There are not enough football basketball games played to have a large enough sampling size to be effective. An additional problem is that everyone is using the same sets of data for analysis. I want to know who is tracking non-traditional data like weights of offensive vs. defense line, etc. Also, the mindset of these kids - who got laid last night, etc. The unknowns that likely make the biggest differences in performance.
Where I have found an edge is reading thoughtful analysis from someone who knows the conference and the teams. Weighting these experts, you can sometimes get upwards of 65% many weeks. These people are few and far between and sometimes need to watch people in the tracker forum to see which conferences they bet. That said, I would love to have another data stream, similar to the "one best bet of the week" that pops up every so often.