Wouldn't part of the point of realtime raytracing would be to avoid a lot of this multipass trickery in the first place and just have a more sophisticated light transport model?
Realtime raytracing hooks into an existing multipass pipeline like this by replacing select traditional techniques. The easiest to replace are shadowmaps and AO. You can also make use of radiance (diffuse GI) and reflections with one or two bounces. Traditional techniques are currently faster and give better results for just about everything else. Also, you can only afford maybe 1-2 raytrace samples per pixel at the moment, even with RTX, so they make heavy use of AI denoising techniques.
I have no idea how RTX actually works, but based on the fact that exploits the existing rasterization architecture, I assume it adds at least one (very fast) pass.
Yes, that seems likely. But can you use it to drop some of the others?
Probably, yes. The only issue I see with that is the fact that Raytracing is only supplementing current rendering methods, not replacing them yet. The performance is not there yet. In a few generations we might see fully raytraced games, but not for a little while I would guess.
Could be wrong here, but I think most uses of raytracing like this limit it to shading and use traditional method for shadows. (Shading being the way that light "appears" on surfaces, shadows determining roughly how well illuminated a surface is. In theory you could use raytracing for both, but there may be limitations?)