the CA's become single-point-of-compromise for the whole PKI. It's a bit more nuanced than that. A compromised CA can issue new certs which can spoof old one's identity, but it can't be used to create certs which will read traffic from an existing host. The big risk is that someone is subject to a MITM attack, doesn't clue in on the certificate change (their client should note this), and accepts the new key.

Signed / validated keys (used for major sites in some browsers) also help this somewhat.

Not that SSL/TLS isn't a huge mess.

A compromised CA in a client cert PKI world, like the DOD one I was talking about, is effectively an end-game scenario. Every machine in the org. would have to be updated (root CA's don't have CRLs, afaik). The big risk isn't a mitm attack - that's a risk with traditional server-side certs. The big risk is that all of a sudden there are unlimited legitimate users with no control over identities anymore. It's like getting write access to /etc/passwd.

Doesn't that depend on how you authorize the client certs themselves?

It's one thing to _issue_ a cert. It's another to _approve_ that issued cert.

If you're blindly accepting certs, you've got other issues.