The large encoded size for small files is due to the overhead from replicated metadata required for storage and authentication in Walrus. This can make the encoded size much larger than the original file size, particularly for small files since the fixed per-blob metadata size is significant relative to the data. For a file like your 16KiB example, this might appear as a 62MiB encoded size when the system assumes a large number of storage nodes.

Walrus is currently more efficient with larger files because the relative overhead for larger blobs decreases as file size increases. For smaller blobs, improvements are being developed, like support for tar balls and random-access reads. In terms of performance for saving small files, sub-second save times might be unrealistic at this point due to the complexity of operations required, such as encoding and sending transactions through Sui and to various storage nodes.