Doom3's id tech 4 lighting engine [2004] trumps HL2E2's source engine [2007] by a long shot. At least the lighting portion.
This is the reason I strictly distinguish between graphics and graphics technology.
The Source engine is built upon compromises.
Smart compromises.
If you want to get the atmosphere of bright day/sunlight environments, you need light bouncing, the difference between
this and
that. The Impressionists of the early 20th century, they already knew. It's not even about hard or soft edges, it's way more complicate than that. And even
Crysis can't handle true global illumination yet (especially not in DX9). Popular graphics cards can't handle this effect in real-time, so Valve, in order to get a lighting feature that, ironically, exists since Quake 1 but
not in Quake 4, they chose to use pre-rendered lighting. And I am thankful for that, because HL2 with pitch-black shadows looks worse than HL2 with static lightmaps and basic, fake dropshadows.
I guarantee that with a coming big update of the Source engine, maybe with Episode 3, we will see true, efficient global illumination in Source. And once they got that, photorealism is close to solved, IMO. I looked at the sky today, I saw no clouds moving, but if they have some time left, they can add dynamic clouds a la Crysis etc. But there are things to be done first.
HL2 sticking to pre-rendered lighting so far was an artistical, a practical decision, a weighting of what they need to create a certain atmosphere and how to get there the fastest. A decision that brings Valve profit (and DAMN fine looking games) where Crytek is struggling. I prefer an Episode 2 map, a TF2 or even a
Portal map over the kitschy, brown and blurry cyber-samurai aesthetic of UT3 or a Crysis map that runs at 15 FPS on a 1200$ computer. Art over technology, performance over technological leadership.
Valve is one of the few companies that, IMO, found the perfect balance between having big ideas and focusing on what's important.
Why snark works.