In this case it'd likely be talking about corrupted, truncated, or otherwise mangled packets. So having to retransmit things. You'd still technically have the

In this case it'd likely be talking about corrupted, truncated, or otherwise mangled packets. So having to retransmit things. You'd still technically have the same bandwidth and latency on the connection itself even if it's causing all kinds of overhead on upper levels of the stack.