Here is a link to the PostgreSQL doc page on MVCC. The choice quote (emphasis mine):
The main advantage to using the MVCC model of concurrency control rather than locking is that in MVCC locks acquired for querying (reading) data do not conflict with locks acquired for writing data, and so reading never blocks writing and writing never blocks reading.
This is why Jeff was so confounded by his deadlocks. A read should never be able to cause them.
