Thats not really true. Assuming your nodes can be ordered, and assuming you store a root id, you grab everything from the tree on or later to the node you are interested in. Then filter out the noncompliant nodes when you read them. One query sufficies then. Admittedly at the cost of some client side processing, but if that isnt a problem then it should be fine.

For the record this exactly what is doe in Recently Added Threads. We grab all the relevent nodes in two queries normally. (Roots then children as our roots arent the same type as the children).