If the requested prec is much smaller than the current size of the pi cache, it could be cheaper to just compute pi from scratch.