Performing an XML backup on a large dataset takes a very very long time and causes an OOM.

Taking an XML backup of a large dataset does not complete. The system spend most of it's time computing the node graph of what to include in the backup.

I've attached a ZIP file with two jprofiler snapshots of the backup. The first is taken from near the beginning of the export and the other from near the end, just before it failed.

For reference, the dataset used is that of CAC.