did you look at pgbouncer ? thats the simple pooler for postgres, and its quite robust, because its so simple.
Yes, it is one of the solutions I do consider. Having applications decide whether they should write to master, or use slaves and/or master for read queries (for instance in case it is a transaction, etc).
I wonder however, how others are handing it. There seems to be nothing out there apart from pgbouncer and pgpool. And only the latter can handle (albeit not really that quick) pooling between master and slaves.