Let me repeat that I'm very impressed by both Kal-El and by NVIDIA (as a company). The possibility to have a quad core ARM Cortex A9 chip and very capable graphic is nothing short of awesome, especially considering that Kal-El is expected to consume less power that the current Tegra2 design! In respect to power efficiency, the aging Core2 has no chance here. Moreover, we must consider that the current Kal-El silicon is a preproduction sample, and so performance can go higher (but also lower) that what NVIDIA anticipated.
However, the published Kal-El vs Core2 results are simply wrong. You can not compare two extremely different chips (and architectures) with different compiler version and different settings... this is meaningless. Moreover, CoreMark is a synthetic benchmark and so, while it is very useful in a broader hardware test, it has little value by itself. So, claiming by only a (wrong) CoreMark score that Kal-El is faster than a Core2 processor is a bit too optimistic to my eyes. The problem is not in CoreMark: it is a very good tool by itself. The point is that it has to be used correctly: if you want to compare two CPU, you had to use the same compiler version and settings.
So, while I'm generally impressed in how fine NVIDIA works, I think that its benchmark division performed poorly in this specific case.