Running Windows servers for high-volume production environments is, if all else is equal, slightly more expensive, and slightly less robust than Linux. I've run both, and seen both run, and have enough data points that I feel capable of making that judgment.

But all things aren't equal. If you know how to run Windows servers, and how to develop in C#, and you have nobody on the team who's a seasoned Linux sysadmin, then you'd just be adding pain and learning curves to your project if you moved away from Windows. The cost in risk to your project by using an unknown platform is likely higher than the gains you'll eventually get (assuming you get to scale and longevity) if you re-tool and re-learn to host on a new platform.

Recommending against MySQL is kind-of silly. Approximately 10 out of the 10 top web sites use MySQL in production, including Google, Amazon, Facebook, and others. It has idiosynchrasies, sure, but so does any other database. And while MS SQL Server has great software development environments for Visual Studio developers, it's not as good with the support for massive horizontal sharding across medium-performance servers, where MySQL shines.

Other good databases to consider include PostgreSQL (a champ in "correctness" while not being slow) and DB/2 (if you need an "Enterprise" database.) Do whatever you can to avoid Oracle. It really isn't worth it for any new project IMO.
And, of course, you could design for a cloud infrastructure, like Google with BigTable, or Amazon with Dynamo/SimpleDB. This means relying on a specific hardware provider, but it gives you a lot of benefits, and the scalability story (should you succeed) is very close to as simple as "push a button."