Large Cubemap Optimization

I'm trying to texture a very big sphere (a planet), and I have encountered this issue: How could the loading of cubemaps (possibly multiple maps-texturemaps / bumpmaps / ...) be optimized to not take up too much of the GPU memory, while beeing able to maintain a decent resolution?

In short: Is there any way to dynamically load parts of a cubemap, or even load multiple textures, manually load / unload in a script and sample them in the shader, the same way as a cubemap is sampled?