And from this, we derive how we should fund this cost using tx fees. We want to keep tx fees artificially high, so that the total cost of mining is high, so that the network is secure.
How do you imagine you will be able to keep transaction fees artificially high?
I can imagine a future with 1MB blocks full of zero-transaction-fee transactions (all fees paid off-blockchain through special cozy relationships between big merchants/exchanges and big miners. Or simply big merchants/exchanges mining their own transactions).
I think network security CAN be funded through transaction fees-- that is easy, if you want to buy some security just attach a larger-than-strictly-necessary-to-get-confirmed fee to your transactions.
I don't think we know yet whether network security WILL be funded through transaction fees; there might be a free-rider problem that keeps people who want a secure network from actually paying for a secure network.
This is where it would be lovely for some academic economists who have studied the free-rider problem to chime in and predict what is likely to happen, and how other markets have solved (or not) the problem.