One thing I can say for sure that last night's fork taught me, is that miner's will earn a greater profit on average if they are able to verify the blocks faster.
Half-baked proof:
Let τ be the average time it takes to verify a new block, let T be the average block time (10 min), let H be the fraction of the network hashrate controlled by the miner, and let Pvalid be the probability than an unverified block is, indeed, valid. Clearly, 0 < Pvalid < 1.
The fraction of time the miner does not know whether the most recent block was valid is clearly τ / T, which means the fraction of the time the miner does know is 1 - τ / T = (T - τ) / T.
For a given block height, assume three outcomes can occur for the miner: (1) he finds a block before he's verified the previous block, (2) he finds a block after he's verified the previous block, (3) he does not find a block. Assume also that if he finds a valid block he receives the block reward*, unless he was mining on an invalid block (in which case he receives nothing).
The expectation value of the miner's revenue is the expected revenue during the time he doesn't know, plus the expected revenue during the time he does know:
= (25 BTC) Pvalid H (τ / T) + (25 BTC) H (T - τ) / T
= (25 BTC) H / T [ T - τ (1 - Pvalid) ]
What this shows is that since the subtracted term, τ (1- Pvalid), is strictly positive, the miner's expectation of revenue, , is maximized if the time to verify the previous block is minimized (i.e., if τ is as small as possible). The limit, as τ -> 0, is
limτ->0 = (25 BTC) H
QED
How does this relate to the blocksize debate?
As the average blocksize gets larger, the time to verify the previous block also gets larger (assuming no other technical innovations). This means that miner profitability begins to depend more heavily (than it does now) on how quickly they can verify blocks. And since mining has thin profit margins, this means that miners will be motivated to improve how quickly their nodes can perform the ECDSA operations needed to verify blocks.
Greg Maxwell once said that he wanted to launch an alt-coin where the proof-of-work was based on ECDSA verify operations. The reason, he said, was the incentivize optimization of ECDSA libraries or even the development of custom ASICs to perform the work. The analysis above suggests we can achieve something similar by increasing the blocksize and allowing normal market competition to do its thing.
*For the sake of this argument, assume the probability of orphaning a valid block is zero (I don't think this assumption affects the current results but it makes the math simpler).