I think theres some misconceptions in your node. My code actually will do less queries than NN. Yes it could profit from some performance tuning, but overall its actually fairly conservative with how it interacts with the DB. Constructing the trees is fairly cheap. Im not going to go into why PM needs so many queries, suffice it to say that there is a fairly good caching system in play so that commonly used nodes are almost always in cache. OTOH, the caching system means every time we "use" a node we have to do version check on the cached value. Since everything is a node in Everything pretty much this means a lot of version checking and because of the nondeterministice way it happens you cant even batch the checks up. So all together we arent reducing the number beyond a marginal amount without totally different architecture.

When and where we see optimization possibilities we do our best to exploit them, but theres only so much we can do in this framework.