Philo

It might be worthwhile to look into ways of making the QuerySetMapper more efficient. As it is, every access of Entity.attributes returns a new QuerySetMapper instance, and each access of a QuerySetMapper instance's keys causes at least one query. (I am fairly certain that, since we're dealing with GFKs, it may in fact be 2 queries.)

In general, instances of models aren't reused excessively - with the possible exception of ContentTypes and other heavily-cached model types. The attributes of a model are reused even less frequently. It might make attributes more efficient if we could cache:

1. key values on QuerySetMapper instances
2. QuerySetMapper instances on model instances.

Another area that could be more efficient is queries on passthroughs. Currently, if x is the number of queries required for checking an Entity's attributes for a given key, the number of queries required for checking a TreeEntity's attributes could be as large as n*x, where n is the number of ancestors of the instance whose attributes are being sought including the instance itself.