authenticating with XMPP ID (jabber address)

> These are mostly performance, not HA solutions. None of these allow
> for simple delegation (if so, please explain how).

Many high availability solutions also improve performance.  An HA solution does not
necessarily mean having a backup server sitting round waiting for the primary server to die.
In fact, most multi-active solutions are more likely to handle crashes correctly because they
need to be aware of other nodes in the common case.

> Again, this does not allow for the XRDS document to be simply delegated
> to someone else (if so, please explain how).

As I said, you can use the same techniques as used to make any web site highly available.

> You make it sound like I only have to introduce it to one part when
> I really have to introduce it to every part for it to actually be HA

It wasn't my intention to make it look like you only needed to add HA to one part.  Some of
the techniques I mentioned can be omitted though (a single high availability OpenID provider
might satisfy your needs, for instance).

> But how can I delegate this to others with openid? The protocol does
> not provide a mechanism to do this? Can a friend easily provide a
> backup openid service for you?  How would the naming scheme work? If
> my identity URL is http://mydomain.com/myopenidsoftware/John.Doe, how
> can my friend's openid server which gives him an identity URL of
> http://backup.com/backupopenidsoftware/Friend use his server to
> backup my id? That is the problem, there is no easy standard way of
> doing this? 

Take a look at the discovery portions of the specification again.  In particular, the
difference between the "claimed identifier" and the "OP-local identifier".  The identifier
allocated to you by an OpenID provider is not necessarily the same as the claimed identifier
given to the relying parties.  Provided you have your discovery information in order, this
should not be a problem.

Now, this isn't to say that there are bugs in the existing OpenID implementations.  Some
deployments are buggy (e.g. SourceForge's implementation is incomplete, with no support for
XRDS).  Few libraries try secondary endpoints in the XRDS if the first fails (this is roughly
equivalent to an MTA only trying the first MX record).  Things will have to improve, but they
are not inherent flaws in the specification.  We are talking about a fairly new technology
here.