Why is X5687 slower than X5570?

[crashulater]

I am wondering what information I am missing that makes the X5687 multi-core performance slower than an X5570 system.
I have two near identical Z800 systems (FF825AV). The first is equipped with two X5570, 2.93GHz CPUs and 12x1GB PC3-10600E memory. The second is equipped with two X5687, 3.60GHz CPUs and 4x4GB (2Rx8) PC3-10600E memory. Both systems are configured "hyper-threading off and all cores on" in the BIOS. Over-clocking is not utilized. I ran an explicit finite element benchmark, similar to the "neon_refined" benchmark, used at topcrunch.org, but not identical. The job is deliberately CPU (floating point) intensive and uses just 2% of memory. The following results were obtained for single and multi-threaded jobs, n=1,4,8:
X5570 n=1: CPU time = 1698 s; Clock time = 1745 s -> speedup = 1.000
X5687 n=1: CPU time = 1575 s; Clock time = 1627 s -> speedup = 1.073
X5570 n=4: CPU time = 2787 s; Clock time = 720 s -> speedup = 2.424
X5687 n=4: CPU time = 3165 s; Clock time = 816 s -> speedup = 2.138
X5570 n=8: CPU time = 5271 s; Clock time = 677 s -> speedup = 2.578
X5687 n=8: CPU time = 6788 s; Clock time = 865 s -> speedup = 2.017
That the single 3.6GHz X5687 with 12MB cache is barely quicker than the 2.93 GHz X5570 with 8MB cache surprised me. That it is slower, as more cores are engaged surprised me even more so. This is a small job though, so n=4 is close to optimal. Obviously there is something about Xeon X-series processors that I don't know. Please, somebody enlighten me, so I can sleep again. 😉

[Repairatrooper]

It is possible that the 12 x1 gb memory modules provide better memory performance than the 4 gb modules used with the faster processor. I realize you state that minimal memory is being used but anything helps.

Memory chart for Z800 http://h20331.www2.hp.com/Hpsub/downloads/Mini_WP_Z800_memory.pdf

[DGroves]

if your test program fits in the cpu with the smaller cache, then it makes no use of the sec cpu's larger cache

however if the app or test can not fit in the smaller cpu's cache then the sec cpu with a larger cache will be quicker