I can buy 100 LTC for a lot less then 40 of these chips
SF3301 chip does have the best power efficiency in both BTC and LTC mining.
LTC power measurement
Voltage(V) Frequency(MHz) Power(W) HashRate(MHahs) W/MHash/s
0.8 450 2.18 1.21 1.79
0.9 700 4.34 1.89 2.30
BTC power measurement
Voltage(V) Frequency(MHz) FBB(V) Power(W) HashRate(GHash) W/GHash/s
0.6 225 0.0 10.98 36.00 0.31
0.8 625 0.0 53.60 100.00 0.54
0.6 300 +- 0.6 14.88 48.00 0.31
0.8 870 +- 0.6 77.68 139.20 0.56
0.6 450 +- 1.1 23.28 72.00 0.32
0.8 950 +- 1.1 97.28 152.00 0.64
Having the best power efficiency means little if the device is priced unreasonably. From this table the most reasonable setting for BTC mining seems to be 72GH/w at an efficiency of 0.32W/GH/s (efficiencies above 0.5W/GH/s have no advantage over current competitors' hardware). Let's compare that with an S5. You'd need 16 sfards chips @72GH/s to achieve 1155 GH/s. Each chip is currently offered for $30, so that's $480 just for the chips. Then you'd need to add the supporting hardware around it; printed circuit boards, a controller, DC-DC converters, supporting circuitry, a case, connectors ... and a markup. How much would such a miner sell for if the chips themselves are $480? The S5 goes for $360 US ...
And in terms of efficiency, what's important is the *system* level efficiency; chip-level efficiency is not that useful beyond academic discussion. Even Spondoolies' current Rockerbox chip can easily be shown to achieve better than 0.4W per GH/s. Put those chips in a practical miner, with supporting circuitry and DC-DC converters, and you quickly see that chip-level efficiencies can no longer be achieved on a system level.
As far as I know, there are no 'practical miners' made with the SF3301 yet. And by 'practical' I'm referring to something that can compete on a GH/$ level with current miners such as the $360USD 1155GH/s S5. Until system-level specs are demonstrated for an sfards-based miner, there's little point in discussing chip-level efficiency.