Michael Lewis tries to analyze Goldman's giant subprime bet (helpfully, he ignores other companies that made similar bets, so he can write a story about how Goldman was unique rather than uncommon). Really, I don't see what's special about this situation, other than the scale: if everyone agreed that there was a 95% probability that mortgages would pay off, we'd either see 95% of traders make a little money or 5% make a lot of money.
But what's crazy about Lewis is not that he has to be dishonest to tell the story he has in mind: it's that his story is so dull. According to Lewis, the problem with Goldman Sachs is that a company with 30,000 people, most hired for their smarts, don't all agree all the time.