I think I see. The argument is that even though attacking ethereum to short-sell ETH may not work (as a successful 51% attack would also burn >1-5m ETH, so it may well on net make the price go up (!!)), but it could easily interfere in any dapps on top of ethereum in the meantime with no compensation. This is certainly an issue. Mitigations I can think of are:
- Designing dapps so that they maximally “fail safe”; try to avoid assuming liveness on scales of less than a few weeks as a security assumption.
- Plasma and state channels - even if the underlying blockchain fails, most layer-2 constructions lead to zero loss of service if you temporarily introduce the additional assumption that the counterparty/operator is honest. That is, for plasma or channel constructions to break, both the main chain and the counterparty/operator must be trying to attack you.
- Protocol changes such as rent and fee reclaiming which burn a portion of txfees; this ensures that the value of ETH goes up under conditions of high demand for dapp usage, making it less likely that dapps will have a high value without ETH having a high value.