Only because the scene is static though. If you wanted good reflections of interactive models on a volumetric blob with a shiny material (eg the T1000 from Terminator 2) in real time you're going to need ray-tracing.
It is not static, the light source moves, it is the same as if the models would move. They explain that they update light probes, they could do the same if the models would move.
I think the limitation here is the number of lights. They use shadowmaps for direct shadows. You can update shadowmaps interactively for a couple of light sources, but if you have dozens of lights, as is often the case with architectural scenes, shadow maps updating can become a bottleneck.
We have precalculated all probes and lightmaps (25) in advance due to defined light trajectory.
This clever hacks, tricks the user into believing it is a real deal - that was a desired goal of demo.
Doing true GI in WebGL wont be commercially viable option, yet.
The Terminator 2 movie is a prime example of using envroinment maps which are supported by OpenGL since day one. There is zero raytracing used in this movie. This shows again that from a psychovisual viewpoint 100% accurate reflections are utterly pointless.
Hardware manufacturers should focus on things that really matter which is texture resolution and geometric detail.
T2 was rendering static environments, not interactive ones in real time. And they used something like 600 SGI computers and months of rendering time to do it. To get that level of graphics in an interactive environment you can't fake it. You need ray-tracing or something similar.
As for texture resolution and geometric detail... improvements there are great as well. It's not a "one or the other" choice.
Not that any of this matters with regard to Playcanvas because there's no way to access the RTX's ray-tracing pipeline from WebGL.
I think the limitation here is the number of lights. They use shadowmaps for direct shadows. You can update shadowmaps interactively for a couple of light sources, but if you have dozens of lights, as is often the case with architectural scenes, shadow maps updating can become a bottleneck.