In fact there is no difference. MeshSmooth and Geo Approx do the same thing on polygons. They both subdivide the object, using the Catmull-Clark algorithm.
What is different is the tool itself. In max it's an operator (Modifier), and in XSI it's a property. Each software offer different options. In XSI you can have your subdivision level based on the distance from the camera.
One thing that I think is missing in XSI is the ability to set a different level of tesselation for rendering and in the viewports. This when the +/- keys are handy. Also, in XSI you can't freeze the Geo Approx, but you have an option to create a new object that will be subdivided (with Duplicate using Geo Approx).
So in the end, when you use MeshSmooth in 3ds max you are actually doing sub-d modeling. It's just labeled differently.
Hope this helps Salutations - Cheers Bernard Lebel
st_petersburg: You are right. Basically sub-d is a concept that allows you to model in low resolution, then to smooth things up when its come to rendering. Well, in practice a low res model for broad cast can be up to 15 000 polygons (or even more), depending on the level of detail you want.