Salt depresses the freezing point of water. When you freeze salty water, the behavior depends on whether you’re at, above, or below the eutectic point, -21.1° and 27% NaCl. If less saline than the eutectic point, first you get ice crystals sucking up heat and growing without much salt until the remaining brine is at the eutectic point, at which point it freezes. If more saline than the eutectic point, instead you get salt crystals sucking up (much less) heat and growing without much water until the remaining brine is at the eutectic point, at which point it freezes.
On the other hand, if you have salty ice and you melt it partway, the part that melts first will be the eutectic phase, leaving the remaining phase (either water or salt) more concentrated in the remaining crystals. At this point you can mechanically separate the two with an arbitrarily small energy input.
I was thinking that you could use this behavior somehow to get a refrigerator, but now I’m not sure you can. It seems like the movement of mass into the eutectic phase is kind of unrecoverable by freezing and melting. Maybe you could separate them with a pressure swing or by distillation.
(This comes from a discussion with SpeedEvil on Freenode.)
But suppose you have a conventional freezer, cooled using a conventional refrigeration approach like the compression-condensation cycle or an ammonia-absorption cycle, and you'd like some resilience against problems --- either power outages, as in Balcony battery, or mechanical failures. A phase-change thermal reservoir inside the freezer is one way to achieve this.
However, just putting water bottles in the freezer is suboptimal. Normally the freezer is at -20° or so, and the water bottles won't melt until 0°, which means that all your food also heats up to 0°, possibly allowing it to spoil or get soggy. A lower-temperature phase-change material would work better, ideally something that melts just above the freezer's normal temperature, like about -18° or so.
You could try using just near-eutectic NaCl salt water, but you run the risk that, with repeated freeze-thaw cycles, it will separate --- I didn't understand this until SpeedEvil explained it to me.
Suppose it's a little lower in salt content than the eutectic 27%. Low-salt-content crystals will freeze first, concentrating the salt in the remaining solution, and at -20°, you will be left with a large amount of lower-salt-content crystals floating on top of the remaining very-nearly-eutectic solution at the bottom. If this system (because "mess" sounds too unappealing) melts, the higher-salt-content water is denser and thus will tend to stay at the bottom, although there will be a small amount of diffusion at the boundary layer. If it then refreezes, the same thing will happen again, but this time the initially-freezing crystals will be even lower in salt content, and thus will freeze at an even higher temperature.
Repeated cycles of this will eventually separate the mixture into a large amount of very-nearly-eutectic mix at the bottom, which never freezes, and a small amount of very-nearly-pure water at the top, which freezes far too easily, separated by a very thin diffusion layer.
I think one possible cure is to make it a little higher in salt content than the eutectic 27%, so that the crystals that freeze first are higher in salt content than the eutectic, concentrating water in the bottom. (I'm assuming that the crystals will still be water crystals and not salt; if not, this won't work!) This way, when the system melts again, the denser, saltier water liberated from the crystals at the top of the solution will not be stable in that position --- it will produce convection cells that carry it back toward the bottom of the tank. This should provide some mixing, but I'm not sure if it will be enough.
Another possible cure, of course, is to use an impeller or a spoon to mix the liquid when there is liquid.
A third possibility is to immobilize the whole solution in some kind of gel, such as agar, so that the crystals cannot float or sink. This way, the physical distance between the different phases of crystals remains small enough to prevent diffusion. Over time, though, I think crystal formation in repeated freeze cycles will cut the gel matrix to shreds.
A fourth possibility is to use a Peltier cooler on top of the salt-water tank inside the freezer to drop it an extra 2° relative to the rest of the freezer. This is a relatively small temperature difference and a small heat flux, so the Peltier cooler should be relatively efficient, although its waste heat does add to the freezer's load.
A fifth possibility is to find another material whose aqueous solution has a slightly warmer eutectic point, one toward the bottom end of the freezer's normal temperature range. There are an abundance of inexpensive and nontoxic salts, acids, and bases that might work. Surely there is a database of their eutectic points somewhere.
A sixth possibility is to just run the freezer a bit cooler, perhaps cooling the brine tank directly and arranging things so that heat from the outer walls diffuses first through the food and then into the brine tank, thus keeping the food a bit warmer than the brine tank.
Solar photovoltaic energy is on track to be by far the cheapest source of energy to date, but energy storage --- especially kilowatt-scale energy storage --- is not cheap. If you can run your freezer only on sunny days, when energy is abundant, you can save yourself or your municipality the need to use scarce cloudy-day or scarcer nighttime energy to keep your food cold. The kind of phase-change thermal reservoir discussed above is about two or four orders of magnitude cheaper than batteries, which are still US$100 to US$200 per kilowatt hour (US$30 to US$60 per megajoule).
Why do I say such a thermal store would be cheaper? According to Thermodynamic systems in housing, water's heat of fusion is 333 kJ/kg, so 3 kg per megajoule --- at the US$0.58/kℓ cost for expensive reverse-osmosis water from Sorek cited in Calculations about desalination in Israel, water costs US$0.0017 per 3 liters and thus per megajoule. Adding a controlled amount of salt might add a bit to that cost, but I don't think it's more than an order of magnitude. According to A minimal-cost diet with adequate nutrition in Argentina in 2017 is US$0.67 per day, supermarket salt cost AR$41.65 per kg in 2012 when AR$16 was US$1, so that's almost US$3 per kg, or US$1 per 3-kg megajoule; but I'm pretty sure bulk rock salt is at least one order of magnitude cheaper than that.
A freezer designed for use in this way would probably be a bit larger than the freezers we currently use.