Primate labs and real performance

As with any synthetic benchmark, Geekbench is just that--synthetic. Which means that it doesn't map directly to real-world performance in a particular application, which might rely more heavily on some particular instruction, or memory bandwidth, or disk speed, or something else. Threading will also make a huge difference--a single threaded process won't get ANY speed boost out of multiple cores, where Geekbench scores reflect an app able to use all of the available processor cores.

That said, Geekbench is a pretty good broad-spectrum synthetic benchmark, so it's going to give you a pretty good idea of the relative speed, from one system to another, of things that are CPU-bound. It'll give a particularly good idea of relative speed difference between systems with the same number of cores.

For example, comparing the currently-available i7 iMac to the new MBPs, you can see that while both are four core, and the iMac is clocked about 700MHz faster, the MBP is scoring about 10% higher, which pretty clearly indicates that Sandy Bridge has a substantial per-clock-cycle advantage over the older i7 architecture. Comparing it to, say, an old single-core G5 iMac, however, it's not going to be eight times faster unless your application is very multi-threaded.