Erasure coding did require more CPU time, that’s true. Still, a reasonable erasure encoding library can generate encoded data at at least 650 MB/s, which is unlikely to be the major throughput bottleneck over a wide-area network with unreliable storage nodes.
– Replication is bad for decentralized storage, part 1: Erasure codes for fun and profit
Yes it needs more CPU time, and we would need to benchmark it and see if the real cost is higher than the gains.