No mean hard-clipping response, but individual grains will only get photo-activated once, which is a sort of elemental clipping.
The same would happen with a one-bit digital sensor; you'd have a linear response right up to the point where the first photosite is hit twice, and can't change any further, and then as exposure continues, more and more photosites are hit and there are fewer and fewer that can still change.
I just did a simulation of a 100 kilopixel sensor that received 10,000 spatially random photons per frame in an animation, turning the canvas eventually from all-white to all-black, and it took about 120 frames before the last white pixel turned black. That means that it took about 12x the mean capacity of each frame to completely saturate it.