First of all, PIR systems currently do not scale well and secondly, the saying "anonymity loves company" definitely holds true even in your brittle single point of failure example design. Consider a PIR system which is only used by Alice and Bob; a network observer can see that the PIR system is only used by Alice and Bob. Although it's true that you do get all of the properties you mention, this isn't good enough.
PIR systems are interesting and I am looking forward to seeing them be practically deployed. This isn't a competition and instead of bashing something different we could be more welcoming of anonymity network diversity. Further, perhaps you didn't consider that mixnets can be used to construct PIR systems... and that we can also use PIR systems in our mix networks as well. Consider the Loopix design where users send messages to destination Providers which queue messages. The retrieval of these messages could be done using a PIR system.
PIR systems are interesting and I am looking forward to seeing them be practically deployed. This isn't a competition and instead of bashing something different we could be more welcoming of anonymity network diversity. Further, perhaps you didn't consider that mixnets can be used to construct PIR systems... and that we can also use PIR systems in our mix networks as well. Consider the Loopix design where users send messages to destination Providers which queue messages. The retrieval of these messages could be done using a PIR system.