Yes, it's possible to have a PUPI collisions. (It's called pseudo-unique, after all, not unique.) This can happen with probability 2-32 if the PUPI is chosen randomly, or more if it isn't. The question is, so what? How is a PUPI collision harmful?
The primary intent of the PUPI is to be unique at a particular point in space-time: an RFID reader emits a “hello y'all” signal (REQA or REQB), and all the cards that receive this signal respond with a message (ATQA or ATQB) containing some type identification information (e.g. “I'm a German ID card”) and a PUPI. The reader may then decide to start a conversation with one of the cards that are in range, and it uses the PUPI to identify which of the cards it will talk to. It's irrelevant whether the reader has talked to different card with the same PUPI in the past, or whether there's another card reporting the same PUPI to a different reader.
If there are two cards in range with the same PUPI, it's likely that the reader won't be able to communicate with either (both would reply if prompted). Having multiple cards in range is supported in principle, but isn't guaranteed to work anyway: it requires the analog signal to be of good enough quality, and depends on how much interference there is (due to opaque or reflecting objects in the vicinity, among others). While a card attempting to trigger PUPI collisions is a DoS, there are worse ways to cause a DoS (it's radio, so it can be jammed).
Another element to the PUPI is privacy: if a card always reports the same PUPI, it's easier to track it. But you can already track a card by type: the PUPI only matters if there are several cards of the same type in the same general location. Ideally, the PUPI should be randomized at each powerup (I don't know if German ID cards to that), or should be identical on all cards (which would also ensure privacy but could cause collisions if there are several ID cards in close proximity — not a very common occurrence in day-to-day situations).