the OP was referring to transfer of the entire block chain so I didn't realize it had abstracted from that to discussion on new blocks with the initial block chain being separate.
He may have been unclear, but it is 84GB "per month". It wasn't about initial download.
144 blocks per day * 30 days per month * 20 MB per block = 84.375 GB per month.
I had meant unless 5000(or more specifically, equivalent to however many nodes there are) new nodes were added, there wouldn't be a 1:1 ratio of transfer of the entire block chain.
Just to stay synced (assuming maximum size blocks), you need to download 84GB. Unless you plan to leech (in the Bit-torrent meaning of the word), you need an upload to download ratio of at least 1.0, which means an upload of at least 84GB too.
If no new nodes join the network, then you are correct, initial download doesn't add anything. Even if the number of nodes stays constant, it is likely that some full nodes will drop out to be replaced by other full nodes. In addition, some (potential) full nodes won't stay online all the time, so they push up the ratio for everyone else.
To contribute to the network a full node needs to stay online and be accessible by SPV clients. Running a full node which doesn't accept incoming connections uses up slots on full nodes which do.
Using the reference client as a wallet means downloading the chain in short bursts and probably having an upload ratio of less than 1.0.
SPV clients are leechers in the pure network sense of the term, but they contribute to the network by increasing use of the currency. They probably won't consume much extra bandwidth by an increased block size.
84GB per month is a lot. This would lead to growth in full nodes or centralization of the system. Do you think all blocks will be 20M? There may be smaller and bigger blocks in the blockchain?