The app I work on is a single server writing to a dozen or so queues. The throughput is pretty low but I still manage to see failures for a few minutes once or twice a week.