I think you miss the point. I´m primarily talking about using those renderers to replace the way lightmaps are created. In the end you will still just get a lightmap that is used exactly as they are used now. The only difference, they are created with a much better method that is way faster, more precise, supports distributed rendering in a way to use 100% of all available power at all the time (unlike lightmass) and enables lighting possibilites that are not to achive with lightmass at the moment. Regarding your comparision of a photo and a painting. One of the reasons I choose Vray as my primary renderer is because it also offers the possibility to render stuff in a not completely physical correct way. For example you can tell a light to affect only certain objects, a mirror to reflect only certain objects or a shader to use different materials for reflections or refractions. Just to name a few examples. Having those abilities are very important when you´re into arch or product viz. If a customer wants a certain effect you can´t tell him, oh well sorry this doesn´t work as it wouldn´t be physically correct. A modern renderer offers you way more abilites to precisely control the look of your scene than lightmass does at the moment. No matter if you want a photo or a “painting” you can do all of this with a modern renderer. You´re stating those Vray or Corona Images are done with unlimited time and unlimited geometric detail. This is pretty much wrong. First of all. You can get decent images that are close in quality with lightmass, no doubt. But it takes ages to render and because of that they are a pain to debug when you have lightleaks or blotches. Actually corona or vray or whatever will safe you a lot of time when rendering lightmaps. If it doesn´t matter for you that a final lightmap calculation takes hours or maybe even days to calculate, well… that´s good for you. But if you want to bring the unreal engine into new industries where you have customers that demand a certain quality to be produced in a certain amount of time, than lightmass is simply not enough. You could get 1000CPUs and lightmap still would not be a single second faster because of the way it handles distributed rendering. By the way lighting quality is not defined by the geometric detail of a scene.
Implementing a realtime pathtracer as it was brought up in this thread and shown by OTOY probabely wouldn´t make much sense for lightmake generation. It would be a nice to have for a quick lighting preview but that´s it. When baking lightmaps you would probabely want a production mode renderer where you set a certain quality with a variable amount of time (wich does not exclude pathtracing methods) rather than a progressive renderer where you would set a fixed amount of rendertime but in return get a variable quality per lightmap. But well I think that´s probabely how OTOY will handle lightmap creation anyway. Also Brigade from OTOY way allready hyped 2 or 3 years back. They don´t have the best reputation in the community. Hopefully they deliver this time.
Using progressive rendering in an application or game as the final GI solution is probabely the future, but a far distant future. As we can see in the example the noise is still way to heavy to get decent results. It will take quite some generations of GPUs until you get 30 noiseless images per second in an indoor scenario without having a nuclear plant and supercomputer in your basement. As we can see for the moment it still takes 1-2 seconds to get a half decent image in a bad resolution. I mean that´s allready fantastic but simply not enough for realtime use. 1-2 seconds means we need hardware that is like 100 or 200 times faster than now. I´ve tested a lot of progressive renderers that are out there. Non is magically faster than the other. They all use roughly the same allgorithms and all have roughly the same speed. It´s just a matter of hardware you have underneath. OTOY afaik uses their GPU render cloud for some of their demos.
But as long as realtime pathtracing is in the distant future I´m more worried about lightmaps 
No matter if you´re doing lightmap baking for games or archviz or whatever. After all it would mean you get a better lighting quality in a shorter time while it doesn´t cost a single millisecond more rendertime in the final applications.
Until some method of realtime GI magically appears, that offers the same quality lightmapping is the best solution and it probabely will be for quite some time. Non of the things I have seen really convinced me that a realtime gi of that quality is is just about to happen. Well, I still wonder why no one ever implemented imperfect shadow mapping into an engine. This was the best realtime GI I have ever seen. Strange, but there might reasons I´m not aware of.