If we're going to use excess power to run compute, I'd rather it go towards protein folding simulation. Something that might actually prove useful.
Every 1 hr of mining at X hash rate will yield a fairly predictable (though fluctuating) $Y of output, but every 1 hr of protein folding will not necessarily yield any direct or predictable income. From the perspective of profit-maximizing energy producers, I suspect they will prefer the former, unless someone can quantify or monetize the latter more easily and directly.