The x{i+1} here is the next x along a given wake -- not the updated position of the same x at the next timestep.
So, u,v,w do not have to go to zero for this to converge. Instead, the wake line needs to lie parallel to the local velocities along a streamline.
That said, it does not converge. Notice that even with many wake iterations, the residuals do not drop to zero. They come down, but then they stop. For simple cases, they don't really improve all that much compared to a fixed wake.
Lift and Drag do a pretty good job of converging, but the wake points wiggle around forever.
Dave is actually working on a mode that will address this for other reasons -- it won't be ready for release for quite some time (I have to get the huge release before that one done first). It is a good bit slower, but it provides much more formal convergence behavior to the wake.
Rob