Soviet Heavy said:
As in a game that uses all of their experiments. Every one of Valve's games so far has been used to promote a new method of experimentation that the devs try out. Half Life 2 ... [yada yada] ... Left 4 Dead experimented with AI behaviours. L4D2 was a challenge set out to see if Valve could release a full game in under a year. Portal was a continuation of physics with humor added this time. ... When do you think Valve will end their testing and release a game that implements all of their little experiments added to their previous games?
I can't accept that L4D2 is a "full game." Sure, they can sell it in a separate box, just like Microsoft does every 3 years - whether their product works or not... uh ... OK, bad analogy. VALVe products are polished, reliable, and properly tested, unlike that company. Portal is alleged to be the MormonTabernacle Drop team (sp?) as assimilated by VALVe, and I have no reason to doubt this. Merely creating the visual innovation that is looking through recursive portals was a staggering accomplishment (from the coding side). If that is the caliber of talent that they recruited, it was a wise addition. In all of your examples, the graphic tweaks, coding innovation, gameplay design, and overall polish are much more complex than "experiments."
Yet you do have a point. VALVe has released a variety of games, each time pushing at least one envelope. My reply to your question is: "I hope they never do."
I haven't played Darksiders, so perhaps I am ill qualified to say so. However, in my experience of cars, its not a good idea to add something cool just because you have the technology lying around. I owned a car that had a brushed aluminum (or at least it was generally silver, hell, maybe it was stainless steel) dashboard. All the way across from the drivers left to the passenger's right. I thought that was pretty damn cool. Right until I had the sun behind me. And reflecting from the dash to get smack in my eyes the whole time I had sun coming through any window. Yep, it sure looked cool, but ultimately a bad idea.
VALVe have done a great job in innovating, polishing, and delivering new user content (TF2). I'll even add to that list something else; showing restraint in adding features to new games. My problem is that I try to take on too much of a project, or too grand of a project. Then I can't finish the darn thing because I'm stuck researching some obscure BS that shouldn't have mattered if I had kept the project manageable. Oh, I'm still with you in wanting HL2EP3 (Manchester 1) released very soon, but so long as they deliver the same tested, polished, quality product, I won't complain very loudly.