I am too cheap to buy Bernstein's book, so I hope someone who has it can explain how they handled the data to come up with an annual alpha of 9%. The best study I have seen of the effect of reporting biases in hedge fund database returns concludes that essentially ALL of the apparent alpha was due to a series of reporting biases. (see Out of the dark: Hedge fund reporting biases and commercial databases, Adam L. Aiken, et al, http://d3gjvvs65ernan.cloudfront.net/Hedge.USA.pdf
Once they accounted for the poor performance of funds once they stopped reporting to databases (but investors' money was still in the funds), the apparent alpha disappeared. Since they, and others, studied hedge fund returns over the same time period as Bernstein, one suspects that the hedge fund alpha reported is before accounting for these biases.
Some, with biased data, have shown an average hedge fund RETURN of 9% (which trailed the S&P 500, but with a low beta), but nothing close to an ALPHA that high.
On the other hand, the databases have been trying to improve their accuracy. In the past, for example, hedge fund managers could put request that the entire history of a fund be removed (presumably when it became inconvenient). As more of these poor returns remain in the database, the apparent average performance goes down.
An alternate interpretation of the result of declining alpha is that the mean return of hedge funds has not changed, but the reported returns are becoming less biased, and so less overestimate the performance.
In short, it is doubtful that hedge funds, as a whole, ever provided anything close to 9% annual alpha. Zero is a more likely figure.
"We don't know how to beat the market on a risk-adjusted basis, and we don't know anyone that does know either."