So with a block period (aka block time) λ of 10 minutes and a propagation time t (for finding a second block) of less than 6 seconds (and propagation will usually be less than 600 milliseconds so that is even a more linear relationship at ¹/₁₀₀₀), then presuming roughly (on average) that doubling the block size doubles both the transaction fees and the propagation time, then the miner has the same income on average with the largest possible block they can make because doubling the risk of another miner finding a block also doubles the miner's income per block statically speaking.
Doubling the block size (and propagation time) won't double your orphaning risk in most cases. It all depends on the block time. For example, if you increase the current block size from 1 to 2mb, your risk of orphaning would only be slightly higher as propagation time would remain well below the 10 minutes block time.
Neither the total fees nor the orphaning risk are proportional to block size (and propagation time).
Well, if everybody is using the default strategy to start mining on the longest chain from the moment of reception, then the systemic orphaning rate IS proportional to the propagation delay, because it is ONLY during this time that a miner might not be aware of a new bloc, finding an old bloc, and propagate it too, making his bloc an orphaned one.
This can be different if people have different strategies, such as selfish mining.
I personally don't believe that the selfish mining strategy makes sense. It makes sense if you also have influence over network propagation delay, and if there is a race condition between your selfish mined bloc and the public bloc, but if every miner is directly connected with a link with EVERY other miner of significance (which is, I think, a mutually optimal design: full mesh between miners of significance), then the selfish mining strategy can't work: you ONLY know that you have to publish your secret chain when you *received* a good bloc from the others, but at that point, ALL OTHERS received it too, and you will be too late with your chain.
"flooding other nodes" doesn't make sense, because of course, you have parallel equipment to each of your co miners' links and the only thing you do is flooding your own gateway, not impeding other gateways from other miners to your peer.
Selfish mining only makes sense in a node network with variable propagation, but that is not the mutually optimal mining configuration, which is direct links between all pools of significance (say, 10, that makes 90 links).
BTW, of course, within the group of 10, they could collude and decide to keep the bloc chain they've made, secret for half an hour or so. Which is the principle of selfish mining. But I don't see the point. They could just systematically orphan all blocks they receive from the outside network too, much easier.