As currently implemented, it appears that if the Primary Identity Authority or Secondary Identity Authority keys are compromised, then all accounts for which that authority is trusted are also compromised. This enables accounts to be compromised en masse, rather than individually. It would seem that by de-centralizing authentication, browserid centralizes security.
Ok, so revoke the compromised certificate. If you "trust" browserid.org and do assertion verification on your own servers, how are you notified of the compromise and key change? Conversely, if you offload assertion checking to browserid.org and they go down, will users really understand why they can't log in and your site is "broken"? You could use CRLs and OSCP, but now you have another thing to get wrong in individual sites' assertion verification implementations.
Another interesting side effect of PIA or SIA key compromise is that an attacker could now create accounts and impersonate users, not just access existing ones.
It would also appear that due to the omission of a nonce, assertions are vulnerable to replay within the validity window.
I think that authentication is broken, and commend attempts to fix it. I'm just not sure that all the added complexity of browserid really buys us anything, and instead enables new attacks that previously weren't feasible.