ModelNode.asPropertyList() results in a clone of the model node. If the node is very large this can be expensive.

We use this call quite a bit in subsystem xml marshallers. The cost is not likely to be significant in most places, but in a few cases where users may end up adding a large number of nodes of a particular type, moving away from this call to another idiom that doesn't involve cloning may help with performance.

I've heard of a user planning to add thousands of datasources, for example, and wanting very fast performance of the write ops to add those.

Things I plan to look at:

(xa-)datasources
messaging destinations
resource adapters
JCA connectors
security domains