Hm, I’m not sure I buy this argument, as it relies on the entire procedure being in equilibrium at all times. For the same reason that the human body can operate at much higher efficiency than a steam locomotive (or even the "perfect" Carnot engine), it should be possible to devise a desalination process which is much more efficient than the thermodynamic optimum by relying on catalysts and other non-equilibrium effects.

> the human body can operate at much higher efficiency than ... the "perfect" Carnot engine

Can you expand on what you mean by this please? The ideal Carnot cycle describes the fundamental thermodynamic maximum for extracting work from heat.

I thought that applied across all state changes (i.e. not just in equilibrium situations)?

Of course it applies if you wish to extract work from heat. However, it does not apply when extracting work from chemical, nuclear or electrical energy, only if you first transformed that other energy into heat and then transformed the heat into work.

People are far less energy efficient than modern locomotives. (In terms of chemical energy to motion.) So, I am now not sure what you mean.

PS: You can also extract energy by mixing salt and fresh water, so the minimum energy to separate them must be greater than that.