Suppose Google says, "We have 78,354 instances of prior art, we'd like you to consider each one in turn. If one of them is good enough, we win. If none are, we lose. Sometime next century."

Since it only takes one piece of prior art for a legitimate defence, it makes sense to say that the defence should only present the best examples.

Limiting the plaintiff does seem a bit harder to justify. If each infringement is legitimate, it seems Oracle should get to try each one. I suppose that if they succeed with these three, they would have a strong case to make saying, "Now we get to try more". So it makes some sense. Try your best examples first, don't waste the court's time and money until you've shown some legitimacy to your claims.