We did some tests benchmarking Opteron box and Xeon box on our MTA system.
Servers: Sun v20z with 2x 270 (dual-core 2.0GHz), IBM x345 with 2x Xeon 2.8GHz with HT.
We switched off second cores in v20z, so we were benchmarking 2x 2GHZ Opteron vs. 2x 2.8GHz Xeon. Both systems were equipped with 2GB RAM. Opteron box had 2x300GB 10K internal disks in mirror, Xeon box had 6x33GB 15K internal disk in raid 10.
System: on both servers Solaris 10 was used. Applications were compiled on Xeon and the same binaries were used on Opteron so no special Opteron optimizations were used. Application is an MTA system with antivirus software, spam filters, and many more modules. As Opteron box has slower disks we anticipated it could be slower, but on the other hand all these applications really need lot of CPU power. We used slamd for benchmarking from several clients.
Result: Opteron server is 1.6x faster then Xeon server. Turning additional cores on and Opteron server is 2.62x faster then Xeon server.
Well, we're going to test our web servers next - I wonder about results.
btw: with all cores turned on we couldn't saturate all CPUs on Opteron server - more client servers were needed and at that time I couldn't spent more (10 client servers were used). Xeon server CPU was saturated all the time. Additionally Opteron had much slower disks, so probably we could get even better results with Opteron.