We're getting pretty close to having the processing power for physics to be a "solved" problem for anything you would really want to do in games, but the real question is what to do with those physical simulations.

Here is a tech demo from 2009:

Computational fluid dynamics is one of the most expensive things to simulate, and I could theoritically build a PC that could deal with some very large and realistic CFD effects in game; mind you, this PC wouldn't be cheap and would (most likely) be a graphics workstation with 2 processors which were both close to top of the line. When the next-next generation begins in (roughly) 2020 and we're using hybrid real-time raytracing and have advanced physics simulations we will be nearly able to create photo-realism with as close to real physics as a person can observe.

The problem (of course) is that producing content for games with that level of detail will be very expensive ... and the new problem will be creating procedural content to make game development affordable.

This.  Games will hit a financial wall long before we'll hit a technical one in regards to simulating real world physics and visuals.

Not necessarily - lot of that increase in production costs has already happened in this generation. If you watched UE4 presentation you could've noticed that one of its main advantages over UE3 is not just improved graphics capabilites, but actually tools and workflow, which allows artist to achieve desired results much faster than today (e.g., global lights don't have to be prebaked anymore). That's why I think development cost increase in this next gen won't be as steep as it was the case in this gen.

As CPUs/GPUs advance I can see more of this trend - e.g., we have lower poly models, tesselation and all the maps (displacement, normal, specular, diffuse...) approach now because current hardware can't deal with original models that artist made in 3DMax/Maya/ZBrush - in some (relatively near or distant) future those models will be just dropped into game editor without need to make all of those steps. Practically, developer will be removing one part of worflow (thus cutting cost) cause hardware will be powerfull enough to handle the original art. Not that I think that production budgets won't grow - they will, but I think that once we hit certain tech level that money will be spent for improvments in other fields.

I see what you're saying, but making a bump map is far quicker and cheaper than creating a 3D model with all those bumps actually on it.

Well, when they model the objects they do it with extreme poly counts - all the "bumps" and geometry is there in original design Than they need to downscale it for game engines (because it "costs" way too much in proccesing cycles to put original), and that's where low poly model and all various maps are coming from. Once we hit enough raw power, original high poly models of object will be directly imported into engine. In both cases artist is making high poly model, it's just in second he's skipping one step (thus reducing time/cost). Though I don't think we'll see this in at least 10 more years.

Extreme poly counts yes, but nobody makes polys with the bump mapping actually on the models as polygons.  Not even 3D animation (such as How to Train your Dragon) used polygons with the scales and hair on the models.  Those were rendered in at post processing.  It would take an impossible amount of time to do that level of detail work on a model by hand.

http://www.youtube.com/watch?v=JYXr7AuPGwE

I cannot express admiration I have for those people and amount of beautifull things they create - ZBrush (industy standard) is actually more like sculpting and not modeling. Once we see that fidelity of models directly in games....

EDIT: Now, day we see this stuff directly in the game....it's rendered using V-Ray, in some future we'll get there in real time...

http://www.zbrushcentral.com/showthread.php?171132-Blur-Studio-Farcry-3-Cinematic-Character-Art

They already use Zbrush to great degree in games.  I mean God of War was made in Zbrush.  I remember looking at the guy modelling the Leviathan, it pretty much looked spot on to what it was in the game.  I understand your point though, we have a ways to go before we get to that level of detail.

Yeah, I know, that's why I linked to them....I tend to go to site from time to time and just go through the galleries and admire work of those people, and wonder when we'll see that level of fidelity in the games...and not necessarily just "photorealistic", real-world stuff - amount of exceptional non real-world and weird models you can find there is amazing, I always find it funny when people are against "photorealistic" graphics, thinking of only usual dark, gritty stuff that got mistakenly associated in this gen with technical "photorealism" which we're striving for.

for racers and maybe shooting games yes, everything else, I'm not so sure. A game should not have ultra realistic graphics, if those graphics are not helping make the game have better gameplay. LA noire would benefit from such graphics, but not Mario in my view. Also, I just feel the detail we get from games now are just right.

The Last of Us is just amazing graphically but it depends on how better graphics benefit the rest of the game as well