Consensus is actually very very cheap if its not done with PoW or similar consensus "tech". Distribution isn't cheap because obviously 2 identical servers cost twice as much as one. But this comparison isn't actually applicable to real world use. A centralized system that wants to have high uptime and reliability does need multiple server too. Most centralized system do not reach the kind of reliability "blockchains" can have because from a cost-benefit point of view it just does not make sense. 99.999% uptime or about 5.5 minutes downtime per year is very expensive and very few public system come anywhere close to this.

>...there's a tradeoff where the cheaper you make it the easier it becomes to attack

That is actually a pseudo-math-security coming form the early days of Bitcoin. Yes, if you double the number of nodes the system is arguably more secure (if the nodes are actually operated by different entities) and it costs twice as much to run BUT there is no point to endlessly increase the number of nodes. Less nodes but strategically placed around the world operated by different entities is way way more efficient and cost way less than the blind "more is better" approach.

Some quick back-of-the-envelope calculation:

1 node hardware + energy and maintenance cost per year $100k USD. 50 such severs distributed all over the world.

This system would therefore cost about 5 million per year. Sounds like a lot but assume there are 1 million transactions per day that would mean 1 Tx cost only $0.0137. Obviously more Tx/d would reduce the price per Tx because operation cost cost does not grow at the same rate.

Give $0.0137 USD to the operators for each Tx sound like a good idea but most system also give the operators the power to change that price, which is comically stupid.

The reason Tx fees on some blockchains are order of magnitudes higher, is because of bad design, completely pointless levels of redundancy, and most importantly because the operators do make huge profits.

Its only logical that they would double the fee if that reduced the number of Tx by less than 50%. They dont care if you cant use the system anymore because the fees are too high for you they only care about the sum of all fees.

Imagine HN but every comment would cost money and the operator would want to make the maximum profit possible. You could plot the average number of comment at a given cost and find the price that leads to the maximum profit. One thing is for sure, comments would not be cheap.