It seems to me that if you add 'rootId' and 'date' fields to the 'replyTo' I used to build the tree here, then you can perform a date based fetch to get all new nodes, run through the fetched nodes to build a list of root nodes, then perform a fetch based on root node(s) to retrieve all the node information you need to build the tree.

That avoids a second table and means you perform between two and the number of root nodes referenced in the time period plus one fetches - depending on how you structure your query.