At your numbers of 100USD/BTC and $0.25/kW
We're actually very close to the point where the cost-at-the-foundry of mining chips meets the cost of solar panels required to power them in areas with good insolation (southwestern USA, for example).
Beyond that point energy efficiency won't matter as much as GH/s/$, at least for those who can get chips direct from the fab and run them unpackaged (and perhaps even uncut in whole wafers) with direct DC power. The solar industry's economies of scale make the blockchain look like somebody's weekend hobby -- they can drive down their costs far faster than chip designers or the fabs can drive up their power efficiency.
Solar puts a cap on the demand for increased GH/J. Even if retail miners won't hit that cap for another year or two (at least) there is at least one organization that has almost reached it already. You'd be amazed to know what proportion of the retail cost of mining hardware comes from having to make it idiot-proof and warranty-safe.
The future is in circuit styles that operate smoothly over the broadest range of power levels. Max-hashrate solar during the day, min-PDP trickle off the grid at night. Batteries probably don't make sense; just use more chips instead of trying to time-shift the energy.
