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.
So if someone infringes a lot of your patents, you only ever get to enforce a small subset? That doesn't make a whole lot of sense.
At some point, Google then releases a new version of Android, where they have changed the parts that were found to infringe so as to not infringe in their opinion, and stop paying royalties.
Is that a new product?
My guess is that for purposes of this kind of order, it would be.
I've always thought there had to be some merit-based or procedural argument to dismiss any part of a case with prejudice. Also, there's cases all the time with more than 3 patent claims at stake, so clearly there's no absolute rule that a case can't go beyond a certain level of complexity.
I hope someone with actual legal knowledge can chime in.