mesh instancing vs combining (unity free)

I've searched on this topic and found alot of information, but I have a little confusion on a couple of key points and was hoping someone could help clear them up. I'm at a point in my project that I have to choose a strategy.

What I think I know:
Instancing
Pros
1 draw call if they share the same material
Can use LOD/Culling (although not unity pro's built in)

Cons
Thousands of game objects? Does this decrease performance?

Combining
Pros
1 draw call
1 game object

Cons
Cannot use LOD/Culling because the mesh is basically always visible
Hard to set up textures/atlasing

Questions I'm not sure about
Does instancing work with unity free?

Do the many game objects affect performance?

Does instancing work from runtime code or just if it was designed in the editor?

I'd love to just instantiate prefabs, but something in my gut is telling me that that can't be the best way because of the high number of objects. With XNA I had created 1000's of objects via code and the performance was absolutely terrible, but XNA is obviously a lot less mature than unity.