In fact, I do believe that perhaps the same Nash equilibrium failure that applies to scripting (as stated above) may apply in the cross-partition design for asset transfers because there is a cascade of history. I need to think about this more. I will try to remember to comment on this point later.
I've touched on this before, but you've reminded me again; partitions are the antithesis of consensus. Taking things to the extreme is helpful to illustrate the problem: with infinite partitions, in bitcoin, you are left with the DAG of the UTXO set, and no blocks or any agreement on what the order of transactions should be, in other words,
no consensus. The LCR in bitcoin constantly forces miners to chose between candidate potential partitions (orphan chains). The nash equilibrium results in rational miners always choosing the longest branch to mine on to maximise their profits.
Talking about partition unification for a moment; if two partitions are totally separate, merging them doesn't have any consequences for ordering because the individual transactions in each partition have been separate from each other, you can order them however you like as long as you obey the parent/child relationship in each partition.