I am very familiar with the basic capabilities of most current database server platforms; it is my business to know where these things top out and what they can do.
Windows is still a mediocre server platform for anything but the low-to-mid range. And even within that range it has some pretty stiff competition. I have yet to see any version of Windows in our data center that has its uptime determined by hardware. Any database platform that can't be fully online 24x7 for at least a year at a time under all circumstances is not a platform that I can use.
Again, I don't blame SQL Server per se. Windows is still not an adequate OS for true mission-critical enterprise database work. I have my complaints about Oracle too, but at least it runs on scalable hardware and bulletproof operating systems.
Incidentally, if I needed to choose an enterprise RDBMS platform that had to be absolutely positively bulletproof and run non-stop for something like 5 years without so much as bouncing a process, it would have to be PostgreSQL on FreeBSD (or even an appropriately conservative Linux kernel if I needed real hardware scalability). Not only is it blazing fast for small to medium loads, it is the most bulletproof and patch free RDBMS I've ever used in a production environment, and I've used most of them.
Oracle needs to watch out for Postgres, it is becoming very good extremely fast, and is feature rich. The next revision (due out soon, or so I've heard) will allow it to scale to very large hardware, including some that SQL Server can't run on, and extreme OLTP loads (as well as a native Windows port I'm told). Mark my words, in five years Postgres will be to the database market what Apache was to the web server market. I was skeptical when first introduced to it since I am accustomed to huge SQL Server and Oracle installations, but have since become delighted with the quality and capability of it. It can't replace Oracle yet for all purposes, but one can definitely see that day out on the horizon.
Windows Datacenter Servers are certified to a specific hardware platform and those boxes approach mainframe level uptimes. Look at the Unisys ES7000 as an example. It has mainframe level redundancy and reliability and scales to 32 CPUs and 96 PCI slots. In testing I am familiar with, it smoked an E10000 in every category.