I did rather mess that calc up didn't I. 
And I can't explain their higher claimed audio power output when the chip is limited to DC power rails of +5vdc and 0vdc. I guess I will have to read the datasheet more careful to see what kind of magic they came up with.
Later:
Ok it's solved, that chip is using a full differential output stage, in effect it is two amplifiers 180 degrees out of phase driving the speaker load between their two outputs, so they can get double the power output that a single stage linear amp stage could wiring the speaker from output to ground. So the theoretical maximum audio output power would use the 5vdc as the peak value (not 2.5v peak) which equals 3.53 volts RMS which equals 3.12 watts into a 4 ohm load. This is again the 100% efficiency figure that the actual chip cannot meet or exceed.
Figure 33 of the datasheet shows this differential output stage with waveforms that should make it more clear how this magic is performed.
Lefty