Matt Asay has a nifty piece up right now about the embedded Linux market. It's well worth reading.
Readers may think, however, that the key to market victory here lies in Taiwan, or maybe even China.
I don't think it does. I think it's much closer to home. (I found this cute little guy at a forum for an Army game. If anyone knows where he lives let me know so I can give a more proper credit.)
In my experience OEMs aren't interested in either design or the cost of their software. They're interested in orders.
Orders come from the companies whose name goes on the outside of the box. From Cisco, say, or Motorola. It's their choices which will make the difference.
In making that choice, I doubt whether they will care much about cost. Amortized over million of units the cost of an operating system is a rounding error.
The key for Linux in the embedded market, I think, lies in its modular design. You can, if you like, use just a few functions from the kernel, and leave the rest. Your "overhead" can be as much or as little as you want.
This still matters. Folks have been putting various versions of Linux on a single chip for years. This has driven Linux' growing market share, which has mainly come at the expense of Real Time Operating Systems (RTOS) like VXWorks from WindRiver.
WindRiver, in fact, has been moving for years toward a more Linux-like design, as cheap chips get more powerful and customers demand more functionality.
There will remain a market for very simple RTOS systems like Salvo but with design constraints falling every year the market is looking to see which "real" OS it will eventually find most flexible.
Right now that looks like Linux.