Gustavo,
The pgpool is an interesting approach to this, but I think that the funcionality of inserting a record at a backend which will be "redirectioned" to other and verifying deadlocks under network demands in acquiring locks on the referenced records/tables in several hosts. Then, IMO, this may be implemented inside dbms. How Marko wrote, this is a non-trivial solution... Really, It could be improved on pgpool to be a process coordinator, but will need some changes in backend too. This is a non trivial implementation, but there are several users waiting for an effective solution for data distributing in a cluster. These users actually buy commercial solutions or build themselves one.
Yeah, I was just thinking that if you start from scratch you're not likely to get very far ... you might look at some of the existing partial solutions (pgPoolII, PostgreSQLForest, ExtenDB, etc.) and see if you can improve them.
--Josh ---------------------------(end of broadcast)--------------------------- TIP 5: don't forget to increase your free space map settings