Interesting question, but I doubt the CPU packaging makes any difference. What could make a difference is the mask version of the CPU and other differences with supporting chips. I am guessing socketed boards are earlier versions of the CPU, so they might actually be not as good since chip production usually improves as the production matures. The board layout probably changed over time to reduce costs and improve reliability with minimal impact to performance.

While doing the test you might also want to swap the ROM to remove that variable. Trying a couple boards of each version would be more statistically significant, I say this since of anyone, mcdermd might have a few of each. I can try running benchmarks on my system with the same software. Which version are you running of Norton?