Answer
Once a write completes, every later read sees it (or something newer), as if there were a single copy of the data. It costs coordination on each operation (extra round trips, a leader or a quorum), so higher latency, and some requests must fail during a partition.