It seems that our update task process is distributed over both application servers and database servers. 20% of the proesses is running on the database server.
OK, here we have the good old terminology mish mash for application servers. The term application server can stand for instances of the application tier of an AS ABAP or for machines.
In your case, you have instances of the ABAP application server running on the database machine and that's where the different byte-orders come from.
When I talk about XML serialization I rather mean CALL TRANSFORMATION ID. But as I said, EXPORT TO DATA BUFFER might serve the same. But the question is, at which point you can make the data platform independent in your scenario, that is serialize into a platform independent container.