In short,
1) fault tolerance: data is stored on more than one computer. A CouchDB 2.0 cluster obviates the need for custom setup of failover CouchDB instances.
2) performance: data is split up and only a part lives on each node in a cluster. That means each node only has a fraction of the work to do (like computing a view index) than a single node instance would have.
3) capacity: with setups of multiple computers storing data, and with data being split among nodes, it is now possible to store amounts of data in CouchDB that exceed the capacity of a single computer many many times, setting CouchDB up for genuine Big Data solutions.