The equity premium puzzle is probably one of the most controversial puzzle in economics. For one, it is rather difficult to measure properly the equity premium, second the puzzle is about a risk aversion parameter that is itself difficult to measure, and third the literature is pretty much all over the place.
Casper van Ewijk, Henri L.F. de Groot and Coos Santing perform a meta-analysis on the topic: they gathered all the papers about the equity premium they could, could the premises and the results and tried to make some sense from all this. They conclude that the equity premium tends to disappear with time and development, and with lower GDP volatility. Thus is should be normal that the equity premium kind of vanished during the Great Moderation. Now get a theory to replicate this. Assume that financial markets develop as time goes, and you can easily obtain a reduction in the equity premium in any sensible model. No need for strange preferences, complex arguments about taxes, or large improbable events.
That said, the equity premium puzzle is about the size of the premium, not how it changes. But given the fact that it seems to be shrinking, the puzzle may soon be moot.