Yeah, there's a common misconception that there's a big bill coming at the end. There isn't - you get to keep the software you used during the 3 years for no charge and you get discounts if you need to buy more.
When building your system during BizSpark, you are unlikely to make architectural choices with respect to software licenses, because they are free. Then you graduate, and you're probably looking at a system that uses way more licenses than it would have if you had to optimize around that parameter from the get go.
BizSpark is an excellent move by Microsoft, and likely is helpful for folks starting out who already know the .NET stack. But I'm very skeptical that working in the Microsoft stack offers such a technical advantage that it is worth the risk that your business is not going to be in a position to suddenly have its operational costs bloated by software licenses by some arbitrary date determined by Microsoft.
It's a tired example, but StackOverflow really is an example of a startup that got a lot done with very few devs or servers and launched a very successful business. Also key is that you can mix and match technologies as you want, something that the StackOverflow used very intelligently (e.g. Microsoft server fronted by non-Microsoft reverse proxy, etc.).
Twitter would have failed had their scalability problems persisted.
In fact, I remember "Six Degrees". It was more or less Facebook, but in the late 90's. IIRC, they ran Cold Fusion on Windows.
Even if you don't have to pay for the first licenses, you'll pay for subsequent upgrades. And designing for the Microsoft makes it harder to migrate to anything else - that's probably why Microsoft invented BizSpark in the first place.
Stackoverflow shows you can get a very significant audience on a half-Microsoft stack, but they have a lot of interesting (as in "different from the ones I have") server management problems because of that. That makes their podcasts much more worth listening.