The 2.6.29 kernel is out

Google do not use fsync. They use triple redundancy for their data and handle single storage node faults as usual in-process working events (i.e. bring the redundancy back to 3, and replace the failed node with a new empty operational one).