It has less to do with current APIs than the fact that whole approach is MSAA takes isn't as suitable with Deferred Shading as it is with Forward. You need to apply further passes/techniques to work around the deficiencies, which is computationally expensive and the end result still is technically not the same as MSAA in a forward lit scenario. A decent example:
https://docs.nvidia.com/gameworks/c.../d3d_samples/antialiaseddeferredrendering.htm
There's little "hacking" in MSAA or "coding it from day1", it's simply a part of the rendering pass just as any other and in real-life development scenario you rarely should use "does this work with MSAA well?" as a base decision-making criteria when factoring the whole rendering pipeline. What matters is does the decision support the overall vision of your visuals. CryEngine 3 had decent MSAA for Crysis 3 for example because the rendering pass itself was pretty well suited to hide it's deficiensies (like spec aliasing) and the fact that it also included a pass for foliage (which isn't necessary traditional MSAA because it can't do alpha properly in deferred context) and still was a very expensive option performance-wise. In fact, you can read up on Crytek's Siggraph presentations that aliasing issues were something they did not fix immediately in the engine when moving from Crysis 2 to 3, but instead found their solution later during the development.
EDIT2: If you want to dig deeper into the reality, this is a good starting point (but for the real journey of development, I recommend going to through the whole Crytek rendering presentation archive):
http://www.crytek.com/download/Sousa_Tiago_Rendering_Technologies_of_Crysis3.pptx
Example of how the problems are solved; in actual development scenario the team will create a multitude of issues and you face a multitude of problems just with wanting MSAA in a deferred context, and you solve them one-by-one until the result is acceptable and you've run out of time, rather than you just deciding to do it properly from day one and then everything will work. That said, if it's a priority it of course should be part of the process, since it's very easy to break and with deferred shading, it's not a simple system in deferred context when striving for quality and performance (unlike forward rendering, where it really is a simple part of the rendering pipeline and most features will naturally work with MSAA).