Given that hash-based distribution allows splitting among an arbitrary power of two number of partitions, with relatively even load, PL/Proxy makes it straightforward to split large independent tables among multiple nodes in a way that allows almost unbounded scalability.
The key word there is independent. If, in fact, your queries commonly cross partition boundaries, the query optimizer will not be able to help you generate good plans for those. PL/Proxy calls can be configured to run on any available node or on every node. If you need data from all of them, that's going to turn into a number of independent queries that need to be assembled with UNION ALL in order to get the full result. This may not execute as efficiently ...
