Matt Mullenweg:

We're releasing the DB class we use for WordPress.com to the world. 
Amoung other things, it supports:

* Replication
* Failover
* Redundant (public/private) networks
* Local and remote datacenters
* Partitioning
* Different tables on different DBs
* Advanced stats for profiling
* More...?

Which is very cool, but it does beg the question: which of those bullet items belong in your data store and not just as a class layered on top of an RDBMS?