Jim Collins has detailed the dangers of superstar CEO in his excellent management book Good to Great.
Now a new study by Ulrike Malmendier of Stanford and Geoffrey Tate of Wharton provides a ton of color and interesting detail around this phenomenon. They find that superstar CEOs tend to have subpar performance (more than would be expected from mere mean-reversion), and that the worst situations are those where governance is weak. These guys are serious - their regressions include a variable for whether the CEO was writing a book at the time!
Someone should cross-reference this with the social/sustainability ratings. I've beaten this to death already, but I'll bet that superstar CEOs underperform on social as well as financial metrics.