Recommended distributor architecture for transactional replication?

Question

We are planning to create a Transactional replication on 2 nodes of SQL Server 2008 R2, each node will be a publisher and a subscriber.

What is the best practice in order to have this scenario done? Microsoft always recommends that the distributor should be independent from the publisher or the subscriber, so it is obvious that a 3rd node is to be involved in the scenario. But, as each node will be at the same time publisher and subscriber, can the same (3rd node) be the distributor for the 2 publishers? Should I create 2 distributors, or can it still be on the same architecture, using only 2 nodes, each one acting as distributor as well as publisher?