I just listened to a talk at the New England Database Summit and learned about a new DBMS that supports extremely high QPS for data in main memory sharded over several nodes with synchronous commit across multiple copies to support HA. Performance is much better when the custom API is used to batch transactions. It has begun to support data on disk.
Now that I think about it, was this a talk about MySQL Cluster?
Note that one does not simply put data on disk. When disk is flash SSD then compression is a really big deal. When disk is much slower than optimizations to reduce IOPS are very important (InnoDB insert buffer, TokuDB fractal tree to reduce random writes).