500mhz raised you only 200-ish points? That's not very linear scaling then. So we can take from that I guess that this test is pretty harsh on a lot more then just the CPU. Since your on a Nehalem generation chip, you're overclocking by bus speed, but did you balance your memory speed back to close to where it was at 4Ghz CPU speed? PCMark is notoriously memory speed sensitive as well.
My benchmark run was with a CPU speed through the roof, but my ram speed is actually sub-stock right now as are it's timings, and my GPU overclock disabled. I was experimenting last week with Civilization V performance and found memory speed to be worthless for Civ V performance. DDR3 1066, DDR3 1333, and DDR3 1600 all performed so closely I couldn't tell the difference and even using the late game benchmark only showed a 1fps difference between each of the above speeds.
Plus, I found that since I'm already pushing my vcore for my CPU pretty hard, I could back off on memory voltage by reducing the ram clocks to 1333 and letting off the timings to all 9's, only costing me a single FPS in Civ V, but in the process allowing me to drop vmem to 1.42v, matching the exact same volts as my vcore (for reference, 1066mhz did not allow any more voltage drops). This SHOULDN'T make any kind of real difference, but it does. Temperatures dropped another 3C on the CPU itself doing this. Thanks to this little trick, I now hit peak temperatures at 5Ghz of only 62C totally loaded up.
But, this trick costs me in artificial benchmarks like this. I'm positive I would break 6K in this test if I cranked it all back up, and may do so later today if I get bored