As to blog/microblog posts, the same thing applies. The communication protocol between clients could specific a mechanism to request a deletion, but there's no guarantee that their client would obey. But there's also no guarantee that the other person hasn't already read your post.
Once you send information out into the world, it's very hard to pull it back in. Even on Facebook, if you broadcast information to people, there's no guarantee that they haven't written it down, exported it, screenshotted it before you deleted it or revoked privilege to it.
The social-service your friends are using will have the same access to the data your friend has access to, and if compromised, you are also compromised. So now you have to trust the integrity of potentially hundreds (actually infinite, since the ability for your friends to carry your authentication for them to access you to a new service would be central) of services - which you can't/don't - and then it's easier to just not trust Facebook.
Where did you draw that conclusion? If you accidentally send the wrong data to the wrong person you've violated your own privacy due to your own mistake.
That's a lot different than you specifying that someone shouldn't get a certain piece of information, but that person gets it anyway because the software erroneously allows them to see it.
It's also a lot different than allowing third parties access to your private data.
Facebook, also, is not just profiles and the feed. How about events, in a decentralized way, for example? Will a single accept will be amplified to the rest of 500 people invited?
For the profile part, my basic identity is my email address. Implement something on the mail server that I can manage and let that be my profile. Emailing a specific command to [email protected] results my profile as the attachment, for example.