It's never been that straightforward, right? According to this paper 20 confirmations is good if you assume attackers can gain 25% of the hashpower, and you define good as <0.1% probability of success.

While this model is supposedly more accurate (I wouldn't be able to tell in a reasonable amount of time), it doesn't differ that much from Satoshi's numbers. The result is directionally identical - the more confirmations, the more guarantee you have that a tx won't be reversed. It does suggest waiting longer than Satoshi's math suggested.