The main reason for saving LibXML instances is that I want to save XML parsing time.

That statement is nonsense. Like I said earlier, the objects need to be serialised no matter what. What reason do you have to believe that serialising to XML would be slower to serialising to whatever you can come up with? Given that XML is very easy to generate and parse, and given that LibXML is a *very* fast parser, serialising to XML might actually be faster! In fact, considering it has direct access to the underlying objects, I'd bet it runs circles around your serialiser and parser. Come back when you have benchmarks.