As alluded to in another thread, I'm building something quite big, and I'd like to make sure I build it as optimally as possible, taking advantage of things like occlusion culling etc.
Traditionally in 3d games you can optimise by using a "caulk" texture (Quake3, not sure about others) to add transparency, when the map is compiled anything with caulk on it is culled, the polygons are completely removed from the scene.
Now, in Second Life this doesn't happen, if you make a surface transparent then it is still rendered, but the alpha of the transparent textures causes it to be invisible. The result is that all those polygons are still there. It therefore occurs to me that making surface that can't be seen 100% transparent doesn't really help, in-fact, a texture with an alpha-layer costs MORE to render than one without.
My alternate strategy then is to apply unseen (hidden) surfaces the "blank" texture, colouring them appropriately to avoid sparklies. It occurs to me then, that these surfaces will be lit, and may therefore incur additional cost in lighting them when they aren't properly visible, so I'm making non-sparkly prone surfaces (i.e - ones that definitely shouldn't be seen) full-bright, so that they are no longer lit.
This method should have the advantage of working better with occlusion culling, as iirc it treats an object as see-through if even one of its face is transparent, regardless of the direction it's facing.
It's hard for me to get accurate results measure the benefit of this method over alternatives. So I'm posting what I'm doing here in the hopes that others may share their experiences optimising builds so I (and others) can see what the best methods are.
One other thing I've found is that bump-maps can cost a surprisingly large amount on slightly older cards, so putting shading onto a texture is generally better, saving bumpiness for surfaces which are shiny but uneven.