it can be done, you could tag each face with an index, and then pass a texture buffer, or array of materials to the shader, and choose the appropiate material.

however, unless your going to be very memory intensive i'd recommend:

A. if the only diffrence in the mesh face's are textures, try to merge the textures(either on-load if you wanna go that far, or more likely pre-process them in some photo editing, and update model uv's to reflect these changes).

B. split the mesh by materials.

I personally employ B at the moment, however the concept of using super textures as of late makes me want to implement A in some form.

no matter what, your vbo is going to have to duplicate vertices for different materials/faces, so if each vertex is given an id along with position/uv/normal data, then you simply need to pass a struct array of your material's as uniforms to the vertex shader, or alternatively, use a uniform/texture buffer and extract the information in them to use.

however there are some things you should note. first of all, older graphics cards don't support tons of uniforms, and your gambling if older cards support texture/uniform buffers(which are standard in i believe OGL 3.3, but are extensions before then.) , secondly you'll probably have to combine this with texture arrays in order to support materials with different texture's. pretty much this would be best suited for more modern graphics card's.

personally i'd make sure you even have to do this first, if your having performance issues, try doing culling techniques to eliminate unseen model's/rooms.

i've never attempted this, so i can't give you any number's/percents on if such a tecnique well actually give you a performance boost, so it's purely speculative from my end.