If i got this patch right, each partitioned table will have the same data structure and store whole rows (it's even more restrictive than previous inheritance mechanism that allowed extending by adding additional columns).

Column or expression should only define in which table an inserted row is supposed to be stored. A single row will never been torn apart. Still it look like a foundation that facilitate sharding BigData(Set) between multiple servers when used in conjunction with foreign data. However a lot of performance improvements will still be needed to compete against solid NoSQL projects (in which you really have a BigData use case).

But looking a bit forward, developing performances improvements on top of an ACID compliant distributed database seems less difficult than to develop a NoSQL project for it to become ACID.