I already explained that cross-partition spending is possible (i.e. won't destroy the Nash equilibrium) only if validation of all partitions is done by all validators, which thus defeats the scaling advantages of making partitions (shards).
You even agreed with this upthread.
I agree in general, but I wonder if there is more to this than it first seems. Yes it is true that merging two partitions into a block will need the block producer to validate both partitions to construct his new block, but that's not to say he necessarily has to keep validating both - if the partitions split apart again afterwards, for example that wouldn't be necessary.
You could formalise this such that an inter partition spend requires the source and destination partitions to merge for the transfer block only, thereby defining a partial order and only incurring an ephemeral increase in validation costs.