Agreed. Also, it looks to me that paper's idea intended to be used for batch (or, at least, batchable) computations, where you submit data, and then retrieve th

Agreed. Also, it looks to me that paper's idea intended to be used for batch (or, at least, batchable) computations, where you submit data, and then retrieve the result. In that case, Wi-Fi isn't used intensively, only for data transmission, meaning it could be idle during computation. So the power savings could be relatively big.