If you're pushing in the flux (which has units of volt-seconds, aka webers) as EMF (the theoretical voltage applied to the space around the coil), then the extra flux goes entirely into leakage: the space around the winding, of the shape made when there is no core present (which, because the incremental permeability drops towards 1, there isn't a core present for that additional flux, so that's why!).
If you're applying that flux to a real inductor (one with finite winding resistance), then the increase in current flow will cause more voltage drop across the resistance, limiting the amount of flux delivered to the actual field.
The drop in permeability is due to the magnetic spins lining up. Additional force cannot make them any more lined up, nor is there a way to spontaneously cause more to appear. There's a fixed population, of limited strength, and that's all you can get.
This is also why ferrimagnetic materials aren't as strong: not all of the spins are in the same direction, in fact some oppose it. This dilutes the population of spins contributing to the net magnetization. They are also usually lower density compounds, as opposed to pure elements or alloys (i.e., some atoms in the crystal are inert). Zinc manganese ferrite ((Zn,Mn)Fe2O4) has a maximum flux density around 0.45T, while (Ni,Zn) ferrite is lower, and YIG is lower still; in contrast, an alloy of iron, nickel and cobalt (100% magnetic atoms) peaks around 2.0T.
You can force the flux density ever higher, but every tesla you add, beyond saturation, is added with the full difficulty of doing it in air alone. That's not to say the core becomes useless -- its ~1T contribution remains present -- but as you push over 3T or so, it quickly becomes space that you'd rather allocate for more copper (and liquid cooling channels!), and so it becomes a matter of efficiency rather than space savings.
Tim