Web-services are supported by a complex software infrastructure that must ensure high performance and availability to the client applications. Web services industry holds a well established platform for performance benchmarking (e.g., TPC-App and SPEC jAppServer2004 benchmarks). In addition, several studies have been published recently by main vendors focusing web services performance. However, as peak performance evaluation has been the main focus, the characterization of the impact of faults in such systems has been largely disregarded. This paper proposes an approach for the evaluation and comparison of performance and recovery time in web services infrastructures. This approach is based on fault injection and is illustrated through a concrete example of benchmarking three alternative software solutions for web services deployment.