sparse files are efficient but they break NFS quota accounting. we run ~10k pods and found that usage reporting drifts and rehydration latency causes weird timeouts. strict ext4 project quotas ended up being more reliable for us.