The primary( and really, only ) recommendation against central authentication is that the provider becomes an extremely juicy attack target. We live in a world

The primary( and really, only ) recommendation against central authentication is that the provider becomes an extremely juicy attack target. We live in a world of asymmetric warfare in which attackers can screw up thousands of times, but defenders only need to screw up once. Dramatically increase the payoff of compromise, and the risk increases accordingly. I know this smacks of security by obscurity, but in the real world, incentives for attackers are absolutely part of the risk calculation.