In the PC world, the OS has become more important than the hardware. If you are buying a server, for example, you first have to decide whether you want to run Windows, Linux, or some other OS. Only after you make this decision can you choose a hardware platform, e.g., Opteron, Sparc, or Power.
The same phenomenon is coming to the DSP world. As DSP applications grow larger and more complex. This makes the role of the OS more important, as it allows the developer to focus on application development instead of minutia like device drivers.
The hardware running DSP applications is also getting more complicated. Again, this makes the OS more important, as creates a layer of abstraction that makes the complexity manageable. For example, an OS is an absolute necessity for a multiprocessor DSP system.
OK, you say, I agree that having a solid OS is important. But who says I have to choose an OS before I pick a processor? For one thing, you limit your OS options as soon as you choose a processor. DSPs are getting more OS choices (just last week, new OSs were announced for the Analog Devices Blackfin and Tensilica Diamond cores), but the OS support for most DSPs is still fairly limited. If you start by choosing the processor, you might not be able to find an OS with the features you want.
Then there is the question of portability. Write your code for a popular OS like VxWorks or Linux, you can switch processors relatively easily. Write your code for TI's DSP/BIOS, however, and switching to a non-TI DSP will take a major re-write.
Choosing the right OS is a complicated process, and I can't do it justice in this blog. (This article and these survey results are great places to start.) But I hope I've motivated you to give this process high priority in your design flow.