Memory-wise high poly models are more efficient than large textures (you need to multiply the texture size by 1.5 when using mipmapping BTW).
However, vertices need to be transformed prior to being displayed on screen. So depending on your CPU / graphics adapter, one or the other solution might be faster.