The answer depends on your situation: You could be a start-up that is more concerned with not blowing tons of money, or you could be in a rush to be done. In both situations, your objective is to find inventory sources that work quickly and then massively scale into those inventory sources. I am going to assume that the goal is to be efficient. In this case, I would start with relatively few inventory pieces and a pretty low bid.

The goal of your experiment is to determine what inventory DOESN'T work while spending as little money as possible. By concentrating your spend on a few pubs, you can get to significant results quickly. If you spread out your spend, you won't generate enough clicks or backend actions per inventory source quickly enough to determine if one is actually working. (In my latter user persona, you can simply spend gobs of money to get significant results on lots of inventory quickly.)

So the moral of the story is that you want to get to results, you have to scale the number of pubs you work with to align with the amount of money you are planning to spend. You would like to, within a day or two, rule out bad inventory and focus on new inventory, or alternately, make massive investments in working inventory to print money for yourself.

I would be happy to do a call to further discuss.