Without any kind of calibration whatsoever, my NIKE+ iPod Nano 7g is surprisingly accurate, given the fact that it only counts vertical acceleration impulses interpreted as steps.
On a 10.2km run (measured with MotionX GPS several times), it showed 9.97km. In other words, it underestimated by distance by 2.3%, which I think is OK considering the technology used. I didn't try it on a treadmill yet, but the "real-world" run was mostly on hard surface and in a flat area, using conventional running shoes.
In a hilly region with some trail sections across the woods, sometimes muddy, including jumps over fallen-down trees, it feels less accurate, but unfortunately I don't have GPS data on this specific path, so I cannot verify.
Next, I'll try on a treadmill. So far, however, I'm quite satisfied with the iPod 7G as a workout companion.
After all, I think counting steps might sometimes give even more conclusive results than measuring the actual distance, since it is hard to capture differences in soil and environment accurately, that would influence your energy consumption. On heavy slopes or in difficult terrain you typically make shorter steps, and the distance may be overestimated a bit, but will still give you an indication about the energy needed.