It appears that the serialization format uses the id of objects to establish parent-child or material associations. This means for example that if two different transform nodes have the same name and therefore the same default id, after serialization parent-child relationships get lost for any mesh that uses the second transform node as a parent.
Is that a correct understanding ? It took me a while to figure this out since I did not expect the name to have a critical function. Perhaps there is documentation ? The loader code appears to look for ids. It is somewhat surprising since there is also the .uniqueId property which could be used for serialization purposes to avoid loss of such information (probably with the understanding that it would not be preserved after loading).