On network load, agree that it's unavoidable - the question is the trade-off between host-level load versus network load. The trade-off always is whether you do a form of data reduction that takes cycles on the host machine (like rsync here - CBT elsewhere which theoretically has more data but puts much less load on the host machine) or whether you transfer raw bytes and do data reduction at the target.
Typically I would side on the "do the work with the local CPU" as snapshots are nearly always done during low utilization times and CPUs are nearly always in excess. Networks and storage is less likely to be underutilized.
