Hi,
some kind of experiment: few days ago I've connected 3 devices for SETI-at_home number crunching:
1) Sony Xperia Z2 smarpthone (ARMv7 Processor rev 1 (v7l), Snapdragon 801 Quad-core 2.3 GHz Krait 400)
2) Samsung Nexus old smartphone (ARMv7 Processor rev 10 (v7l), TI OMAP 4460, Dual-core 1.2 GHz Cortex-A9)
3) Acer TravelMate-6292 old laptop (Intel(R) Core(TM)2 Duo CPU T7300 @ 2.00GHz [Family 6 Model 15 Stepping 10], 2 cores)

Smartphones are connected through USB watt-meter between power supply and a phone and laptop is connected through the watt-meter between power supply and a power socket.

Below is a result of Total Credits, Consumed Energy (kWt*h) and the ratio: Total_Credits / Consumed_Energy:

Device               Total Credits     Consumed Energy, kWt*h    Ratio
Sony Xperia Z2       1088              0.332                     3277.11
Samsung Nexus        180               0.092                     1956.52
TravelMate 6292      605               2.78                      217.63 

Is such Ratio a valid way to measure energy efficiency of devices? Watt-meters of smartphones are connected after power-supply. Even if efficiency of power supply is 0.5 (I doubt about it), it looks like smartphones are more efficient devices than the old laptop.

Seems like a good comparison, assuming you calculated Watt-hours properly (5V times current times hours). Your USB power supplies are probably 80+ percent efficient. You could check with a kill-a-watt meter or similar device that would tell you the power drawn at the AC plug. Your power going into the phone is not exactly the power being used to process data - the difference is going into charging/discharging the battery - you might run the test long enough to be sure you aren't charging the battery (or discharging it).

I believe the numbers because ARM is a pretty efficient processor (and why Intel has had trouble getting into mobile device market), and your laptop is pretty old.

My USB watt-meters do measure exact watts*hours independently to current and volt, so it should be Ok. My tests are based on 4 days measurement, but I agree about batteries, it will be good to remove them (could be problematic for Xperia Z2).

So, it looks like buying relatively modern smartphones with broken touch-display (as my Xperia Z2) is a good way to build the power-efficient BOINC computation rack. Perhaps much cheaper than making it from the multiple Raspberry-PIs.

P.S. Why Android applications are not used GPUs ?