I had exactly the same reaction when I got to that part of the letter. I think you can take any trend and extrapolate it into incredulity -- Buffett's career is certainly evidence of this.
This is taking the index numbers at face value. Better assessments of returns using good inflation numbers and incorporating tax liabilities make the picture much worse. http://www.itulip.com/realdow.htm
Finally, the real sample size goes back much farther than 1900. We have good equity price data back into the 18th century. 80% market crashes were common. Overall returns were not good.
Stocks were cheap in 1980. Then they soared in price for 20 years. Now people are slow to realize that period was exceptional. Investing is about "buy low, sell high" not "stick your money in stocks." Stocks are high. The smart money bought commodities in the early 70s, changed to bonds in the early 80s, changed to stocks in the early 90s, and changed back to commodities in the early 2000s. That smart investor has absolutely crushed an equity fixated investor. This isn't rocket science, either. It's not that hard to review asset classes once a decade and figure out what's very cheap and what's expensive.
Where will it flow next? Probably wherever it isn't this time.
Right now commodities (a.k.a. "stuff") is getting the attention, as some have predicted for years (someone's always predicting something for years, until it comes true).
Recently: Tech -> Housing -> Stuff -> ?
A problem for people with money to worry about. ;)
I will say that these trends seem to have followed a 17 year period (on average) during the 20th century. If this remains true america has another 9 years of slow economic growth (and commodities have another 9 years of bull). But we must always beware of the induction fallacy :)