What's the tradeoff here? This just seems like the best way to do it otherwise
Well I think this is the best way. Another way is to let the RGB sensor collect data separately with 3 channels for one pixel. In that case, the resolution would be lower and there's a slight offset.
I think the production cost would be a matter here. Each layer capture only one of the color thus require three layers of of sensors could potentially triples the cost of production although one layer is only responsible for one of the colors. I guess it would be the problem of $/pixel
I'm also a bit confused about why this hasn't been iterated on more, this company was acquired in 2008. Has there really been no way to innovate more here?
From what I've read, the benefits are not as clear cut as they first seem. To start, there's issues with noise and cross talk, of which a better explanation can be found here, not to mention the lack of Adobe support (you have to use Sigma's software), which severely hinders postprocessing.