It does switch between the two automatically. I know on my Ti4200 that they were separate (because the 3d part failed whilst the 2d part worked, so I could use windows but not run any games), I dunno what the newer architectures do. I suspect they're all integrated.
I believe that's because it's unnecessary to overclock a card while in 2D, thus increasing heat (and possibly fan speed). Why bother consuming that little bit more power and making that little bit more heat when its unnecessary?
the ti4200 didn't support dynamic frequency switching; you need a FX at least to make use of that.
Note: it is also accessible through Nvidia's driver when using the coolbits registry hack. It is disabled by default on desktop GPUs, but enabled on mobile ones.
Geforce4 and older: no clock switching (neither CPU nor Video RAM)
GeforceFX: GPU switching
Geforce6: GPU switching
Geforce7: GPU and VRAM switching (? please confirm)
The card detects if you use 3D elements (textures, triangles) and switches to its 3D clock in under 3 seconds (this happens when using Direct3D as a video output too) then switches back to 2D clock 5 seconds (on average) after no more 3D elements are displayed.