Yes, in general, to it being a good idea. OpenGL generally has transparent loss mechanisms, that is, if a texture is lost, you won't notice it.
The idea of making multiple textures into one large texture has been explored. This is called "Texture Atlasing". There are faults with it if you try to make an atlas that is greater than the maximum texture size of the graphics card (which on modern cards is quite large, 8Kx8K or so).
There are pathological cases where texture atlasing will not improve performance, but with discrete graphics cards offering >> 512MB of video memory on low end cards, I don't think you'll run into any of these type problems. I wouldn't concern yourself too much with performance optimizations if you are developing for the desktop systems until it is clear that they are needed -- and what an epic title it must be then! ;)