Let p be the percentage of water and m the total mass.
Then m(p) = 1/(1-p) . The "unintuitive" aspect is that we want to think of this function as linear when it isn't.
By taylor's theorem, how bad a linear approximation to this function will be is based on its higher order derivatives. We can get an idea of this by looking at its second derivative:
m''(p) = 2/(1-p)^3. If you plot this graph, you'll see that it really starts to blow up past 0.8, so the nonlinearities start dominating.