Linux is only one of many systems that are providing innovative ideas in the area of operating systems. There are some people that would have Linux (or NetBSD, or OpenBSD, or, horror of horrors, Microsoft Windows NT) be the only operating system in use.
While I certainly do favor seeing Linux in wide-spread use, and think there is some value to seeing it in universal use nearly everywhere, I do not favor the idea of making it, or any other system, the exclusive choice. While I may be monotheistic in spiritual matters, I believe in pluralism in the context of computing.
There are many reasons for exclusivity to be a bad thing.
It has often been bad for system security:
99% of viruses target the ubiquitous MS-DOS and Microsoft Windows. In a more "pluralistic" OS society, such viruses would likely only hit some of the systems some of the time. Similarly, the Internet Worm targeted SunOS and Ultrix machines; other systems on the Internet were immune because the code simply couldn't run on them.
With multiple platforms to be targetted, virus writers are more limited in the sorts of damage that they can do.
Typical approaches to breaking security involve targeting specific "boundary conditions" of the system. Having multiple OSes is likely to result in multiple sets of boundary conditions. Again, finding a security hole in a particular OS will not mean that all systems on the network are therefore made insecure. (Which is the case in all-Microsoft shops.)
This also extends beyond the OS into various subsystems such as news and mail. Sendmail has traditionally been a source of largely platform-independent security holes, thus the presence of a variety of alternative mail transport agents (such as Qmail ), news servers, and web server software limits the amount of damage any one security hole can cause.
Almost as critical is the notion that not all operating systems are good for all things. Embedded applications have significantly different needs from "desktop computing" which have different needs from database servers... It may be possible to run many of these sorts of applications on the same sort of hardware/OS platform, but efficiency may not be guaranteed. There's a place for minimalist systems such as QNX as well as the "richer" Linux environment as well as commercial Unix es that provide support for high availability hardware.
A more interesting reason for a "pluralistic" approach is that by taking different approaches to operating system design, we get a rich set of ideas about computing that encourage innovation.
Indeed, most of the "free" OSes have borrowed liberally from concepts discovered with other OSes and from related projects. FreeBSD has borrowed Linux device driver code; Linux (and virtually all TCP/IP implementations, both free and commercial) has borrowed networking code from BSD; most free OSes of all stripes have liberally made use of the FSF set of tools. The point here is that each of the "free" OS projects has benefited greatly from this sharing of information.
Multiple OSes permits exploration of multiple approaches to improvements.
It does appear that Unix has a particularly useful set of abstractions; a large proportion new OSes being targeted at "general purpose" computing have, of late, built on layers that look like Unix.
Perhaps the nicest web page summarizing work on operating systems on the Web is Current Operating System Projects and OS-related research at the University of Arizona. It does rather a nice job of categorizing them, and provides what I would regard as an extremely relevant set of one paragraph briefs characterizing each OS.
The page Operating Systems on the Web presents another view, as does Brad Appleton's Operating Systems Links.
The Google Directory of Open Source Operating Systems provides yet another interesting list of operating system.