You have to define what sort of accuracy you mean. Relative or absolute. A better question might be which is more useful? I am going to go ahead and invite flaming to say that even F1 wind tunnels are quite inaccurate in absolute terms. You can test the same model in two tunnels and get very different results, in fact I have actually never seen it be the same or frankly very close. Repeatability is always much more important than accuracy.
CFD is still less accurate for everyone that I know. It is usually more cost effective.