After reading this thread, I am starting to wonder if I have always worked backwards? :smileywink:
I always start with a blocky shape and work my way up from there. As I work up, I save the objects as the different LOD levels until I get to the highest LOD, where I am happy with the shape and number of polys/tris. But still keeping it low. At a low LOD level I also create the UV-map.
I then "upload" or at least just go to the upload window to see the numbers.Then I will know if I can go higher or have to go lower on the tri count.
The less verts I have to move and work with the easier and quicker it is.
(You = the general "you")
A good practice for learning to optimize your objects is to look around you and draw, and I mean use pen and paper.. , objects you see around you, but draw them as blocky as you can. Then do the same thing in your 3D-program. If there is a change in the surface of the object, look at it from different angles, how far away will you see it in-game (SL), do you really need to add extra triangles for it or can you add the details in your texture map?
And like Tarius said, you don't need any extra polygons on a flat surface. If you're used to working with sculpties, you need to have polys on flat surfaces because of the UV-map, but with mesh you're creating your own UV-map and a flat surface might only need one poly (2tris).
Please note, it's the end result that matters, how you get there is not important, each person has his or her own work flow.