The fact that federated identity connects together the current archipelago of user's identity islands is sometimes presented as enabling (or at least exacerbating) identity theft. The connections are seen as amplifying the consequences of any breach, e.g. a domino effect where one after another of your accounts is compromised.
But, if federated means connected, the different accounts of many users are already implicitly federated through the passwords that they reuse to access those accounts, this percentage of users reported as high as 40% (I confess I do it for "disposeable" accounts). If one account is successfully phished and the password is stolen for that account, it's a fair bet that that user's accounts at other providers will be accessible with the same password.
In addition to making strong and different passwords more useable for the end users, federated identity makes the connections between islands explicit and thereby manageable and controlled.
But, if two currently implicity federated (through shared password) accounts are explicity federated (through SAML 2.0, ID-FF, etc) and the passwords stay the same, then the risk of the implicit federation will remain. I'm sure users would love to be prompted with a request/demand to make sure that the passwords used at the two providers were different.