The ray-tracing process is converged from left to right. The view dependent part is detected and only such pixels are ray-traced. The rendering of the view independent part keeps more than several frames par second (typically we keep up to 14 frame/sec). And a perceptual function teaches to the ray-tracer which pixel should be important and traced first. In this image, the closest glass is detected as important, then, the ray-tracer shoot rays to the glass first.
Next mng animation shows a sequence of upper images. I hope you can see this animation.