Could it be because we've been adding 6 figure grid references, which have an accuracy of 100m. But the lat/long would be taking that reference literally, so it would plot it at exactly that point, though the stone could be like 50m? away. Which is why they're not all out by a certain amount in a certain direction.
And presumably the only way round it is if we visit every site and take a gps with us. Off you go then.