This problem is usually considered from the other angle: if variance is high but the odds are unfavorable you should favor fewer bigger bets over many smaller bets. Why? Because the more bets you make the more likely you are to converge onto the expectation. A single big bet leaves you either running way over expectation or slightly under it. All of this assumes that you have to bet though, the safest thing to do with negative expectation is to just not bet.
His example is unusual in that it has positive expectation (so you want to bet) and high variance (because you’re limited to a single bet). f(x) = 0b or 10000b so the single sample has a mean of 5000 and a variance of 50,000,000 but the E[f(x)] is 7500. A bigger than Kelly bet is rational here but if you had unlimited (or a large number of) bets you should revert to Kelly.
Food for thought.