I can provide some insight into this argument, I am a game developer!
So, this is what it ultimately boils down to...
Years ago when D3D (would eventually become DX) and OpenGL (OGL from here on out) were competing for the API crown, M$ made D3D/DX proprietary, and OGL was an open source standard API.
Now, what M$ did was basically say, windows games can use D3D or OGL, however, non windows games cannot use D3D, only OGL.
If anyone has ever actually used OGL, they know, it's archaic. The API is a mess with legacy code and other things coming along making things confusing for new developers to use. On top of that, OGL doesn't have one easiest or fastest way to do something. You can typically do the same thing 5 different ways, no single way is the fastest, and all are easiest to the guy that did them that way. However, they're confusing to anyone coming behind them trying to figure out exactly what it was that was done; unless they think like the first guy that did it.
Now, D3D has easy to use developer tools, with far more first and third party support. It's not as easy as other APIs, like Glide on 3DFX cards back in the day, however, it's uniform and easily repeatable. This means it's easier to develop on, plus it works in Windows which is the lion's share of DT PCs anyway.
Now, you could make a windows game to run a Linux version pretty easily if you recompiled and used OGL as your 3D API; however, since it's so convoluted and poorly supported, no one bothers to do this. Throw in that DX11 has some features that are more cutting edge, and you can quickly see why many developers just use DX and basically say "screw off" to OSX and Linux in general.
Now, what WINE does essentially is not emulate windows to get these games to work. Instead, it runs them in a virtual machine and basically puts a "wrapper" on the D3D calls it has "tricked" the game into running, and converts them to OGL draw calls.
The complication runs into areas when OGL does not support features that DX does. Up to DX9, OGL has full support, so given enough time and effort, you could theoretically convert any DX9 game into a WINE compatible game. Now, with DX11 and newer, there are some features that the Khronos Group has not bothered to bring to OGL because they are too busy dropping the ball to be bothered (which is typical, they have played drop the ball many times over the years).
Now, fast forward to Steam and all this hoopla about Linux gaming now...
Well, Source Engine and several other engines have been ported to Linux, so you could feasibly build a AAA title on Linux, though few developers want to because of limited market share. For some, the additional trouble to run it on Linux and Windows would be amazingly productive; for other developers, they frankly could care less...
There are, however, currently a list of engines waiting to be ported with proper support.
For example, CryENGINE 3.4 will run on WINE, however, CE 3.5 will not...yet. CryTek is trying to get someone to convert the engine to Linux, but it's a fairly daunting task.
Next up: AMD's new MANTLE API.
This, seriously, could be the massive equalizer. It's an open source standard for the hardware, and allows developers to code much closer to the metal than either D3D or OGL. Plus, it uses the same shader language as DX (HLSL) so you can code for both at the same time.
Now, if this takes off, and gains wide support and acceptance. Then the effort barrier to produce AAA titles on Linux will effectively have the bar for entry lowered dramatically.
So, as a game developer, here's hoping MANTLE takes off like a rocket. Additionally, if you would like to see better future support for gaming on Linux, then help promote things like getting CE 3.5 converted to Linux, getting MANTLE into mainstream games, and/or getting someone to get the Khronos Group off their duffs to rewrite OGL into something at all usable.