Two identical Dell R710 hosts running VMWare ESXi 5, purchased at same time, with identical configurations.

Two identical VM's (cloned - clean VM's nothing on them but Boinc 7.0.28 x32) running Windows XP SP4 with 8 cores (2vcpu * 4). VMDK's on the same datastore on the SAN, etc.

Both systems only in the WorldGrid project running identical tasks.
The performance of one system is 50%+ greater than the other and I cannot tell why.

Virtual Center shows total CPU usage on one at average of 20000mhz+ (peak of 22k), but the other never goes past 12k and many times is around 10k.

Ran the CPU benchmarks on both VMs and they show very different results as well and as expected the higher performance system shows 2826/3843 MIPS the other 1768 /3640. Test run when there is no other VM load on the system, both BOINC clients using identical configs (Run always @ 100%).

Connect remotely, suspended activity and then swap the hosts - and the performance swaps as well - which confirms that the issue would not appear to be in the cloned VMs but something to do with the hosts.

Virtual Center confirms identical CPUs, RAM - everything. As mentioned above, these systems are configured identically. They both contain dual Xeon x5650 @ 2.67 with 96GB of RAM.

Any suggestions on why two seemingly identical systems return such different results?

As you concluded yourself the difference in performance is not BOINC- or VM-related. It could be some bad RAM or a heating issue. I would suggest you do some hardware tests with a Dell tool like the Unified Server Configurator.
Or read more at http://www.dell.com/support/troubleshooting/us/en/nlbsdt1/Product/poweredge-r710

More Proof of differences between Windoze and Linux? These are MAJOR and caused by the benchmarks being way different I think. Also, the benchmarks are fairly unstable and can vary a few hundred points. This may be related to your dual systems differences as well. In fact, I wonder if some precision internal clocks or something inside the hardware could have an effect.

These are project results from running Optima@home: (Linux a bit longer too!)

1100T box running Linux 64b:

1048420 779092 20 Mar 2013 6:28:36 UTC 20 Mar 2013 6:47:38 UTC Completed and validated 1,118.93 13.80 13.80
1048419 779091 20 Mar 2013 6:28:36 UTC 20 Mar 2013 6:47:15 UTC Completed and validated 1,100.07 13.57 13.57
1048398 779070 20 Mar 2013 6:28:11 UTC 20 Mar 2013 6:47:15 UTC Completed and validated 1,119.29 13.80 13.80

Same 1100T box running Win-7 64b:

1042715 773387 20 Mar 2013 2:33:13 UTC 20 Mar 2013 4:57:47 UTC Completed and validated 1,062.06 7.76 7.76
1042678 773350 20 Mar 2013 2:33:13 UTC 20 Mar 2013 4:57:59 UTC Completed and validated 1,062.33 7.76 7.76
1042677 773349 20 Mar 2013 2:33:13 UTC 20 Mar 2013 4:57:59 UTC Completed and validated 1,062.92 7.76 7.76

No difference in hardware... just the OS...

More reason to make all tasks a fixed value as other projects have done since the benchmarking/points system in BOINC is flawed...

8-)