The real world performance you get will vary depending on your usage patterns. Do you run applications that use multiple cores or do they bottleneck on a single core? Are your typical applications disk intensive, in which case the hard drive performance is much more critical than CPU speed? Benchmarks measure different things, so you should look at multiple sources of information.
You can find a variety of benchmark results for many different systems at spec.org. The SPEC folks create the benchmarks, which are typically run by the system builders (Dell, HP, etc.). The results are searchable on spec.org so you can compare to your heart's content.
For CPU comparisons, the SPEC CPU2006 benchmark is available in integer (CINT2006) and floating point (CFP2006) versions, and in speed and throughput versions. Speed measures the performance of a single core, while throughput measures a multi-core workload. I generally use the CINT2006 throughput benchmark results to estimate CPU performance of my multi-threaded server application.
SPEC also provides benchmarks to simulate a variety of applications such as web servers, Java client/server apps, SIP, etc. You can choose the benchmark that most closely matches your application and get a comparison that considers the entire system performance, not just CPU speed.
