Yeah, you're right Hans!
But with the shorter shutter speed the compensation has to occur faster.
So even when you "shake" say 10 pixels at 50mm in 1/50 sec you'll end up 10 pixels at 500mm in 1/500 sec. So you have to shift the sensor 10times as fast for the same amount of pixels.
And that's the real challenge for body based anti-shake
That's not completely true.
Yes the time is faster, but the rotation to be compensated is less.
Let's say the rotation during 1/50 s is phi, then the movement at 10 m is 10phi and at 100 m (tele) it will be 100phi. During 1/500 s the rotation is 1/10phi. So 10 times as fast, but also 1/10 of the rotation.
Actually there will not be a fixed rotation, but a random movement (rotation) back and forth. If the IS is gyroscopically based, it would actually work better with fast movements in stead of slow movements.
Anyway, my point is: don't write that lens based IS works better, unless there are tests which confirm this.
I'm looking forward to the tests with the E-510 and a much longer lens.