These are great points. I think your analysis of fast withdrawals is spot on.
1& 2 are indeed difficult to work around. For 1), the only way to shorten the liquidity lock is by shortening the withdrawal time & thereby increasing the online assumption / increasing the need for watch towers. Dan Robinson has suggested withdrawal times of less than 3 days–which could be considered. 2) is also troubling & fast withdrawals may only make sense for erc20 tokens with relatively high liquidity [Update: I’m much more bullish on fast withdrawals & think that they will generally provide minimal value loss during a transfer from one chain to another].
Agreed that 3) is the easiest as there is a built-in incentive for operators to buy withdrawals to provide users a better UX. However, I think longterm the solution to 1, 2, & 3 is avoiding withdrawals all together except in exceptional cases.
Ideally users will seldom need to withdraw from plasma. However, to achieve this plasma must be: 1) general purpose enough to support a broad variety of use cases, and 2) depositing into a plasma chain does not limit you to that chain’s plapps or liquidity.
With predicates we can support a large number of use cases, and with native cross chain transactions it won’t matter which plasma chain/operator you choose–plapps are not isolated to a single chain & liquidity is shared.
In its ultimate form you could almost think of depositing a coin into plasma as depositing a coin into Eth2. It’s a new scalable trusted enviornment. You shouldn’t want to leave the land of scalability unless something goes really wrong (eg. an operator is malicious). Still, if you really want to, you have the option. And only then will you need a fast withdrawal.