On the facts the above desciption doesn't describe behavior in the network at the moment-- e.g. backlog isn't there, and substantial amounts of the funds that went into the DOS attack paid for outputs, not fees.
The scheme you describe is scale-free; I see you clarified in later messages that you think the "solution" is dyanmic controls rather than a removal of limit but the bold increase blocksize response in your initial is quite confusing-- more than half your text is spent spinning hyperbole, it would have been much more useful to spend that text on describing what you're actually talking about.
Perhaps most importantly, it does not make a case that the attacker produces increased income relative to his hashpower. Consider:
Lets imagine your mine with half hashpower. Lets imagine that a block can contain 6000 transactions. Attacker has 1/2 hashpower. Offered load is 4000 tx/block.
Attacker crafts 2000/tx block at 1coin/tx fee level. Making the rest match him (plus episilon, which we'll disregard).
His average cost for spam is 1000 coin/block (2000 * 1-rate).
His average income is 2000 coin/block (4000 * rate). (He doesn't get income from his spam, he saves its cost however; see prior line)
His net income is 1000 coins/block, on average.
Now consider the consolidation of other miners:
Their average cost for spam is 0.
Their average income is 3000 coin/block (6000 * (1-rate)).
Their net income is 3000 coin/block.
Both groups have 50% hashrate, so the non-attacking miner has a fee income of three times greater the attacking miner per unit hashrate!
Normalized for hashrate thats 2000 vs 6000.
----
Lets instead imagine that there is also a backlog of fees episilon beflow the attackers floor, and he mines those instead of his own and that doing this doesn't somehow eliminate the floor effect:
Attacker average cost for spam is 1000 coin/block (2000*1-rate)
Attacker income is 3000 coin/block (6000 * rate)
Attacker net income is slightly under 2000 coins/block, on average.
Honest miners cost for spam 0.
Honest miners income is 3000 coins/block (6000*(1-rate))
Their net income is 3000 coins/block.
Again doesn't work.
----
We can work this for any other size, say and attacker with 40%:
Attacker cost for spam is 1200 coin/block (2000*(1-rate))
Attacker income is 1600 coin/block (4000*rate)
Attacker net income is 400 coin/block
Honest miners cost for spam is 0 coin/block
Honest miner income is 3600 coin/block (6000*(1-rate))
Honest miner net income is 3600 coin/block.
Normalized for rate, thats 1000 vs 6000.
---
Finally, we already know that the system is not incentive compatible when a single party (or collaborating conspiracy) has more than 1/3rd of the hashrate:
http://arxiv.org/abs/1311.0243 (The results below 1/3rd require information asymetry advantages which are handwavy, but at 1/3rd or beyond no such asymetry is required)-- though such attacks are highly conspicious.
At best, this attack allows a sizable minority group of miners to engage in price fixing without running out of money, under the constraint that legitimate transactions are still wiling to pay enough to fill half the block.
Exactly, like anyone they can generate transactions to drive up fees; large miner hashpower gets a discount on fees; but they still lose funds; and everyone else shares the income.
