Dan and I have spent many, many hours desk-check and testing, including pounding for many hours in DBT-2 at high concurrency factors on a 16 processor box. In doing so, we found and fixed a few issues. Neither of us is aware of any bugs or deficiencies remaining, even after a solid day of pounding in DBT-2, other than the failure to extend any new functionality to hot standby.
At Heikki's suggestion I have included a test to throw an error on an attempt to switch to serializable mode during recovery. Anything more to address that issue can be a small follow-up patch -- probably mainly a few notes in the docs.