Set an arbitrary limit which is way above what we need right now, but closes the attack vector.
agreed.
and last I heard its exactly how the attack remains mitigated in classic...
AS BU supporter though, we don't need limits!
IMHO the financial incentives are strong enough that block size (in terms of both bandwidth to transmit/ and CPU to process) is self limiting. Propagation time is a combination of the two things and to (over)simplify propagation time vs orphan risk is enough to make sure miners don't do stupid things, unless they want to lose money.
The full math is here - David you would probably be interested in this if you haven't already seen it.
http://www.bitcoinunlimited.info/resources/1txn.pdfThe paper also describes how the sigops attack is mitigated through miners simply mining 1tx blocks whilst validating then pushing that out to other miners whilst they are still validating the 'poison' block. Rational miners will validate the smaller block, and they also be able to mine another block on top of this, orphaning the poison block.
The attacker would get one shot, and would quickly be shut out. If you have enough hash rate to be mining blocks yourself its really much more profitable to behave!