If the area of the floor of the swimming pool is increased from A to 2A (analogous to adding the initially-discharged capacitor) then the height of the water (analogous to the voltage across the capacitors) must be halved from h to h/2. The gravitational potential energy of the water (analogous to the electrical potential energy of the capacitors) must be halved, too. The weight of the water (analogous to the charge on the capacitors) does not change when the floor area (analogous to the total capacitance) is doubled.
Indeed if we perform this experiment, using a pool with one wall able to move freely yet sealed water tight, we get a piston, which as it's moved back and forth, in the steady state, bears a force of F = ρ g l h^2 / 2 (for water height h, gravitational acceleration g, water density ρ, and wall length l). Thus we do work on it when pushing (makes pool smaller, taller), or vice versa. Going between the x and 2x cases (for pool width x, 2x), this provides precisely the energy required to make up for the apparent halving of energy while conserving charge. Energy of the system (pool AND actuator) is conserved, exactly as we should expect; but considering too limited of a subset (i.e. the pool by itself), energy isn't conserved, it's been moved in/out of that boundary somehow.
We also have the dynamic case, though an overly messy example of it (Navier-Stokes fluid equations are a bitch!). Suppose we partition a pool in half, filling only one half; then suddenly remove the partition. The water sloshes into the formerly-unoccupied side, and continues to slosh back and forth until friction has absorbed the "AC" energy. Indeed superfluids exist, so we could perform this experiment with such, and demonstrate an apparent perpetual motion machine, where the mean free surface level gives the charge-conserved and half-energy figure, while the AC component (gravity waves on the free surface) continues, in motion, with exactly half the initial energy of the system.
I think, because of the nonlinearity of gravity waves, the (potentially?) biphasic nature of superfluids, the free surface moving in gas rather than pure vacuum (the only known superfluids exist at very low temperatures and fairly low pressures, indeed being cooled by evaporation), there will be too many kinds of dispersion (i.e. one long, blocky wave breaks up into numerous higher order waves), friction (due to the gas in the chamber itself, and the liquid's normal phase component) and other effects, dissipating the wave energy in a real superfluid experiment; but even so, again we have the argument: where is the energy BEFORE it's been dissipated to heat, or equivalently, exchanged into other forms of energy besides where it was? And again the answer is clear, it's stored in both static (or average/DC) and dynamic (AC) modes.
Tim