I've done this a few times. Depending on the size/density of your data, it can be fine. If you have billions of entities and/or highly dense relations, it's not very efficient though.

I didn't use table inheritance, and I don't distinguish relationships at the query stage, so if that's core to your method you should think about it. I assume with that setup it'd be easier to take advantage of partial indexes though.

Recursive queries work but if the intermediate tables gets large it explodes fast. You can't use aggregates in the recursive function either (e.g. to count number of leaf nodes in a tree), you have to apply them at the end, in which case the intermediate table has to be large...

I've had decent experience so far with GIN indexes and @> operators on an ARRAY[] column adjacency list. In my case I stored "ancestors" so I could select by object id and get all ancestors, or use `ancestors @> ARRAY[parent_id]` to select all object ids as descendants.

Of course, if you don't need any "self" relationships, then none of the really complicated stuff matters...