One thing to keep in mind when talking about image stabilization:
In-camera image stabilization typically relies on accelerometers, which shift the CCD around to compensate for some types of shake. Accelerometers are semiconductor components that have a finite responce time. You then need some high-tech logic, which also has a finite responce time, to move your CCD around based on the input from the accelerometers. The advantage here is that every lens you put on a camera with built-in image stabilization is therefore image-stabilized as well to a degree.
In-lens image stabilization relies on spinning gyroscopes. This is a very VERY simple technique that relies on the tendency of a spinning object to maintain its plane of rotation. If Canon and Nikon are smart, they have implemented it 100% mechanically, increasing the effectiveness of the image-stabilizer. I haven't taken a lens apart to find out what it is they are doing, though. The downside is that, being part of the lens rather than part of the camera, you have to buy image stabilization with every lens you get whenever you want it.
I personally think in-lens image stabilization tends to do a better job as it is a lot simpler and can (depending on implementation) potentially have a much shorter reaction time - the one thing to limit performance of a gyroscope image stabilizer is mechanical friction. Today digital logic is regarded as superior to its low-tech implementations, but this is one example where it can never perform as fast or as well as a simple low-tech implementation. Truth be told digital technology is prevalent just because it's cheaper to make on a large scale.
It is up to you to choose the image stabilization system you want. Keep in mind that you cannot have two image-stabilizers working at the same time as together they cannot synchronize and will overcompensate for any shake. This in turn will reintroduce shake all over again, rendering them both ineffective.