For 1000 tx/s with the current broadcast system you'd need 5MB/s upload bandwidth: ~ 50Mbps upload link.
I have a 100 Mbps internet connection (upload and download) for $20/month, but we're going off topic here.
We aren't really off topic when discussing how many full nodes will be affordable in the future: your computations rely on the network having 100 000 full nodes. My opinion is that the number of full nodes will adapt to the cost/benefit ratio of running one.
How much more expensive is an unspendable output compared to a spendable output to the global network?
~$0.10.
If I'm correct with only 1000 full nodes this drops to $0.001. This is why I think the previous discussion about the number of full nodes is important.
It is obviously a flaw in the bitcoin protocol that miners can include a $0.10 network cost transaction with a cost of 0.00008 BTC for themselves.
It isn't if you consider that running a full node will have to be linked to mining in the future. Such a transaction doesn't cost them 0.00008 BTC, it pays them whatever they want to include it in the block. Fees will simply have to adapt to whatever rule the miners put in place, the full nodes without miners behind them don't count at all as they don't bring any valuable service (copying data around isn't much of a service).
And that 0.00008 BTC cost decreases with better connectivity and infrastructure, to a point it even becomes profitable for well connected miners to spam the network (See retep's post). The problem doesn't even fix itself when BTC is worth $1000, because miners are getting the tx fees, not the network full nodes.
This is why I disagree: if running a full node doesn't bring any income to its owner and doesn't bring any security benefit by mining what exactly is it good for?