Artillery is building a hardcore RTS with HTML5 and WebGL

For starters, yes there are complex problems with lots of edge cases indeed. But not one of the many problems you suggest are computationally difficult. Now there are some computationally difficult problems to deal with in an RTS but those are not among them.

Additionally, a large amount of the problems you just listed are common among any game with AI. RTS pathfinding is hard, but so is pathfinding an an FPS, and then add in how much more complicated a level is in a game such as Assassin's Creed, to the more simplified structures you can manage in an RTS, and it's easy to see why this isn't simply a matter of RTS games being simply more difficult.

And within an RTS there very many problems that are much easier, both from a complexity, and computationally then in other games. Visibility determination is one thing in particular that's considered super difficult, and very performance sensitive, that's very easy to solve with the fixed perspective of an RTS.

And then there is networking, in which case you can use a lock-step model in RTS due to being less sensitive to latency, that both vastly simplifies networking. And also makes it possible to synchronize the large amount of units in an RTS with minimal bandwidth.

The problem of WebGL is simply it has very low penetration. By my research, currently only 35% of desktop users can use WebGL with GPU acceleration (FYI, Flash 10: 95%, Flash 11: 75%; Flash 11(Stage3D with GPU): 70%). Now that IE11 is supporting WebGL, this might be a matter of time. However I foresee it would take 5 years or so to achieve 80%.

And, note that WebGL only has functions of OpenGL ES 2.0. Currently WebGL2, which has functions of OpenGL ES 3.0 is being developed. On the other hand, native games already utilize OpenGL 4.x. What API is used in native games when WebGL2 becomes popular? I imagine raster-based rendering might be obsolete at that time.

And I think people already know the web can't handle Oculus Rift, probably one of the most interesting, innovative and fun gadget in this decade. Of course you can use it on browsers with a NPAPI plugin. Oh, hey, the web people said plugin is obsolete and evil! Don't use plugins!!! Don't play with Oculus Rift, you, PLEASE!!!

I admit the web platform has some value on some points, but if people think HTML5 is the (only or most) cutting-edge and innovative technology, that's not correct. Absolutely not.

However it has taken some time to optimize the code to run fast enough on all these devices. The main issues I've come across are normal game algorithm issues (using a quadtree to find nearby minions, using hashmaps instead of arrays for some data that is always searched by id etc) and garbage collection. Garbage collection has been an annoying issue which I've somewhat solved by using entity pools. So instead of creating new entities and deleting them when they die a pool of entities is created at the beginning. Then upon creation each entity is pulled from that pool and after it dies all its variables is reset back to zero and it is placed back in the pool to be used again later. This uses up a lot more memory but it means the game doesn't have a large CPU spike every few seconds as it does garbage collection.

Canvas performance may be an issue once I turn it into an iPhone app as I've heard the canvas app renderer has far worse performance than the browser renderer (I have no idea why), but I'll cross that bridge when I get to it.

So in summary I think it's fine for simple casual 2D games that don't have complex logic (using A* algorithms to move large groups of units may be too much, or having advanced AI). But anything designed to appeal to hardcore gamers who need 60+ FPS and want in depth gameplay may be a bit much for it for now.

Simple - I do think the simplicity of the web is is its greatest strength - it does limit what you can do with the web, but it also means it is not too hard to either produce content or create a browser (though nowadays the browser part is pretty challenging). CSS gradients are a good example in fact of the messy web process of experimentation, consensus, and eventual convergence - you will soon be able to use the prefixless style and not worry about several prefixed versions. Personally I'm not sure prefixes are a good idea anyway, but they do go away eventually.

Open - I meant as in anyone can contribute to the standard, and any vendor can write a browser, no one party controls it, as they do native platforms. I'm quite happy Java is blocked given who controls Java :) On the server of course, you can use whatever you want, whatever framework, language and paradigm you like, which is quite liberating when compared to say creating apps for iOS.

Extensible - well the standard actually allows for other languages, and JS being the only scripting language is more a quirk of history than anything else - there have been moves to allow byte code instead (Nacl), and I suspect that's the way things are heading - to a sandboxed VM which will run any compiled language. When that happens then a huge range of languages will open up, just like server side.

That being said, I don't see why non-locked-down native platforms are bad.

The biggest reason they're bad from my point of view is that they don't cater as the web does to hardware not invented yet (as the web does), obsolete hardware, or to other hardware platforms and are locked-in to a particular chip and instruction set, even if not locked down to an API. The other disadvantage is that you are locked in to whatever API (and typically language) the platform has chosen - everything must be written with that language, or with glue code interfacing to that language.

That's the big difference with the web - it defines a limited reference platform (the browser) which can be made for any hardware, and anyone can target using any server side tech they want. There is a clear division between client and server, and both can change completely without invalidating the other. That's a very powerful abstraction and one which I think gives it an advantage as a platform against any native platform which rivals it.