A poster on the Sparkfun forums (donnib) clued me into the Cirrus Logic CS5460A Power/Energy IC. This part appears to be designed specifically to be the backbone of a power meter. It takes as inputs the instantaneous voltage and current and generates real power measurements as well as instantaneous power.
The datasheet can be found here: http://www.cirrus.com/en/products/cs5460a.html?prodKey=CS5460A
This has many advantages. All the power equations are done for you. Much of the signal conditioning is also done for you. The simultaneous sampling of V and I is done for you. It has a simple SPI interface to a micro with an interrupt line to allow for non polled operation. What is not to like?
To start, the part is SMT only. It comes in a 24L SSOP package only making a bit incontinent for prototyping. The data sheet also has some layout constraints that might be difficult to address without making custom boards.
But after my first reading, the worst part is that the current and voltage sensor inputs are in the range of +- 250 millivolts. Not a problem for the voltage, but current transformers are normally in the range of +- 1 volt full scale. This would mean having to attenuate the signal before going into the part. That seems daft. Take your signal, lose precision during the attenuation, and then have the part amplify the signal and introduce more error. Perhaps I have read this wrong.
At this point, I am thinking that the Arduino approach at 60x over sample would be the simplest approach. Or perhaps interfacing a pair of ADCs to the Arduino to get the simultaneous sampling.
More experiments and reading to follow.