I can definitely understand the points both you and kano made and agree with you that the addition of empty blocks to the chain is a necessary evil in the current state of things.
The issue wasn't whether it's okay to generate empty blocks or not. The issue is a pool creating empty blocks when there are transactions pending due to a pool's poor optimisation choices. P2pool by itself does not do anything to limit transactions as an optimisation, it just uses whatever it is given by the bitcoin daemon. That points the finger at the custom antpool implementation as being responsible.
Yes, it does. Just like the other example kano pointed out. My initial statement was that I felt the addition of empty blocks to the blockchain just for the sake of adding a block seemed to be contrary to what the blockchain itself is: a record of transactions. Both kano and syke provided very good reasons why they are allowed. I accept their points as valid, which is why I stated the addition of the empty blocks is a necessary evil.
I also fully agree that some pool implementations are just plain bad coding decisions. Again, as both you and kano point out, there are a number of pool options that do not make such choices (your pool and p2pool being two such examples). And, just to stay on topic, AntPool is an example of poor implementation since it is producing those empty blocks in the first place.