I think looking at transactions is a reasonable metric. In fact, even this metric may be overstating things. Tim Swanson has analyzed transactions excluding popular addresses (apparently this excludes things like SatoshiDice) to identify transactions that are positive-sum in an economic sense and then the numbers look even lower. [1]

Another factor to consider is that if you count purchasing power transferred (which I assume means dollar value times transactions) you'll end up giving a lot of weight to events like when coinbase/mtgox moved their coins from one wallet to another.

[1] http://www.ofnumbers.com/2014/06/23/separating-activity-from...