For ray tracing like this, going the supersampled antialiasing route is pretty much equivalent to rendering the scene at higher resolutions. For example, to get 4 samples per pixel, you need to render at double the resolution in both directions, or 4 times the number of pixels. Of course, techniques like FXAA which are commonly used in rasterization based realtime rendering are possible as well, and would have much lower performance impact.
Another thing to think about is that in any monte carlo based rendering algorithm, such as what would be used to render modern movie graphics, antialiasing is basically free. These algorithms already work by running thousands upon thousands of samples per pixel, so distributing these samples inside each pixel has almost no performance impact.
Another thing to think about is that in any monte carlo based rendering algorithm, such as what would be used to render modern movie graphics, antialiasing is basically free. These algorithms already work by running thousands upon thousands of samples per pixel, so distributing these samples inside each pixel has almost no performance impact.