Then each data facility could reasonably process a few thousand letters a day, batching them for the postal service. After a few rounds of mixing, if there was significant facility-to-facility traffic, it would become impractical to find any specific letter's path.
Time for real world implementations of our high-latency packet routing algorithms?