@alkan, thanks for that. I hadn't bothered to even look at BU, so that is first time I've seen that white paper. I just got back from running an errand so I will read that paper and respond soon.
You're welcome.
The point being that yes the risk of another miner winning the block increases, but the miner's income commensurately (proportionally) also increases, so statistically the miner loses nothing by creating a larger block and thus is leaving tranactions fees on the table for some other miner to take if the miner doesn't make a larger block. However presuming some transactions pay less per byte than others (and higher valued transactions can afford to pay more per byte), the economic converse effect occurs wherein the miner has the incentive to make the smallest block possible or below the size where propagation latency is linearly proportional to block size (i.e. the latency that is a constant factor independent of data transferred), which is again not a free market limit on block size and not a fee market.
I'm not sure if I got everything right, so please correct me. Here's my take on the problem (off the top of my head):
- For small-sized blocks (like a few 10kb), connection latency is more important as a limiting factor than bandwidth. So, up to a certain block size, propagation time is sublinear to block size. Or more precisely, propagation time gets asymptotically linear with high block sizes.
- The transactions waiting in the mempool of a miner will usually come with different transaction fees (per byte).
- As a result, building a block with twice as many transactions (ordered by fee per byte, starting from the highest) won't give the miner twice the fees, but considerably less.
- Furthermore, getting more frequent but smaller fee rewards from mining blocks tends to be more profitable as you can reinvest the profits earlier (similar to compound interests). This effect is negligible for big mining farms, but can be important for low-scale miners with only a tiny fraction of the total hash power.
- For all these reasons, miners will have a clear incentive to build blocks as small as possible, or small enough so that bandwidth gets negligible as a limiting factor for propagation time.
- When some of the miners start building small blocks, your own incentive to build small blocks will be enforced due to the higher competition.
- The average block size will at any time converge to a (relatively low) value which gives the miners the highest transaction fees.
- Block size will eventually depend on the network's bandwidth and thus follow Nielsen's Law. However, it will be independent from demand!
- As long as demand exceeds supply, there will be a "healthy" fee market, though this market comes at the price of payment congestion and affects the overall throughput and usability of the network.
- If the demand falls below supply, users could set the fees arbitrarily low as the miners would still include the transactions in their (small) blocks, which results in a Tragedy of the Commons.
- Profitability of mining falls sharply and impairs the network's security due to the decreased hash rate. This also encourages centralization.
EDIT: I completely ommitted that doubling the block size won't double the orphaning risk (as long as the propagation time is significantly smaller than block time). So, my reasoning is not correct.