Like everyone else, I simply want to know for myself how Buffett determines "intrinsic value." The two most important clues to me are 1) his reference of John Burr Williams' "value equation" in Berkshire's 1992 Shareholder Letter (i.e., the perpetuity equation) and 2) Buffett's actual use of the perpetuity equation in an example to determine the intrinsic value of a newspaper in the 1991 Shareholder Letter. These two facts lead me to the inescapable conclusion that Buffett does indeed use the perpetuity equation to determine a company's "intrinsic value." Furthermore, the fact that he uses 10% as his discount rate in the newspaper examples indicates to me that 10% is the appropriate discount rate to use. The newspaper example is not merely a hypothetical example: the newspaper's 6% growth rate isn't hypothetical but real, which indicates to me that the 10% discount rate Buffett uses isn't hypothetical either.
Your point about not being able to use the perpetuity formula when g is greater than r is well-taken, however the case never arises that a company's g will be greater than the discount rate, r, forever into perpetuity. A company's g may be greater than r for a time being but eventually it will fall below r. The reason is because of a company's life cycle, wherein a company will experience a growth or "exponential" stage, followed by a slower "maturity" phase. For such companies, a "two-stage" dividend discount model is appropriate, the first stage corresponding to the company's exponential phase, wherin g > r, and the second stage corresponding to its maturity phase, wherein g < r. Buffett himself often talks about Berkshire not being able to grow at 15% forever. He cites Carl Sagan's example of bacteria: though they could double each 20 minutes forever into the future, they don't because the world (and bacterial substrate) is finite. Likewise, any company's positive-NPV projects are finite. Coke will reach its "maturity" phase when every man, woman, and child on Earth drinks 10 Cokes a day (like I do).
Excerpt from finance textbook about dividend discount model (you'll find a similar paragraph or two in every finance textbook you look at): "The constant growth DDM is valid only when g is less than r. If dividends were expected to grow forever at a rate faster than r, the value of the stock would be infinite. If an analyst derives an estimate of g that is greater than r, that growth rate must be unsustainable in the long run. The appropriate valuation model to use in this case is a multistage (read: two-stage) DDM." So you see, g may be greater than r in the short-run -- but not in the long run.