I thought I'd expand on that last sentence because localInstanceOfObject seems to confuse multiple people.
localInstanceOfObject() actually returns a fault from the new editing context for the passed in object.
If you have 2 peer contexts:
databaseContext -> ec1 -> ec2
and you call localInstanceOfObject(), and the object is in the database, you'll get a copy of the databaseContext snapshot of the object when the fault resolves from ec2. That is, the editing context goes up one level to the database context and returns whatever the database context has, not whatever ec1 has (because they're peers). If its an uncommitted object, then the object only exists in ec1, so the fault can't resolve into ec2. Similarly, any uncommitted changes in general from ec1 won't show up in ec2, because they're peers.
What Mark just said was that if you have an uncommitted object, and you have:
databaseContext -> ec1 -> ec2 -> ec3
(i.e. ec2 and ec3 are child editing contexts of ec1, but peers of each other)
Then when the fault fires in ec3, it goes up one level, pulls the objects and changes from ec1 (committed or not) and returns it to you in ec3. But that only works because ec1 is common to both ec2 and ec3. In other word, localInstanceOfObject() returns an object from "one level above" the source editing context which is not necessarily a copy of the object you want, it depends on how the destination context lies with respect to the source context.