Problem is there is no way to prove the majority chain has seen a transaction since it refuses to acknowledge it. The minority chain could include the censored transaction, but the majority chain could include more new transactions. There is ambiguity over which chain is the truth and so the Longest Chain Rule (LCR) kicks in and the majority PoW chain is the winner.
The chain selection rule is the problem here. You'd need a new one to deal with this - it would somehow need to take into account the greater set of all transactions; since censorship necessarily is exclusive,
the true set of all transactions would always be larger than the censored set, such that the minority would always have the ability to include the attackers transactions along with the censored ones. However, quite how you combine this with the LCR is unclear at worst and maybe impossible at best, because the minority by definition has less hashing power than the majority.
The bolded phrase can't be measured. With block chains we don't know when is when. All we know is the longest chain and what was confirmed in each block. As you know, there is no global clock to compare chains. This insight is fundamental to understanding block chains at the research level.
As I explained in the post of mine you quoted from, the only way to measure relative time between two block chains is during propagation, but you and I already documented in my vaporcoin's thread that propagation doesn't prove anything (can be inconsistent among peer nodes and offline nodes can be lied to).