Well, there's diminishing returns in format shifting. The encoded barcode contains various types of quasi-redundant visual information (e.g. error correction codes) to allow decoding to happen, so for audio-based transfer it'd be better to skip the image encode and blast out the file directly.

That said... given the somewhat remarkable way fountain codes work, there's nothing stopping us from having a protocol that uses the audio and the video channels simultaneously for better throughput...