The first report looks at the behavior of Tomcat on both platforms when resource limited, a situation that can happen when a server has too little memory, or Tomcat hasn't been configured to handle the requested load. Surprisingly the behavior was quite different on the two platforms, which is unexpected since the source code is essentially the same. When the server was pressed to capacity, our Windows installation was forced turn away some traffic with minimal alteration in serviced performance, whereas our Linux installation elected to service nearly all connections at the cost of introducing latency. However, prior to reaching capacity, our Linux server appeared on average to be capable of servicing connections at a slightly faster rate than our Windows server.
The next report looked at the performance of Tomcat on the two platforms when not resource limited, and there Linux was able to handle about 32% more users than Windows on identical hardware and identical test conditions. Note that the two platforms still exhibited the characteristics seen in the previous test, that of turning away connections vs. increasing latency.
For more details refer to the complete report at http://webperformance.com/library/reports"