The benchmark goal uses call/1, which interprets its argument at runtime. Thus, instead of measuring the performance of a call, it's measuring the performance of compiling a goal plus the performance of the goal itself.

There's a compiler optimization (when a static predicate is declared but not defined) that prevents the complementing solution from working in the current Logtalk version. I'm looking into it.

The familytreev2 can be simplified by declaring that it implements the family protocol.

Also, if the intention is to have the familytreev2 object as a singleton, the definitions of the male/1, female/1, and parent/2 predicates can be optimized by calling the family/1 predicate locally instead of sending a message to self.

In the benchmarks/dcg_example.lgt file, there's no need to send a message to self to call the predicate in the included file. simply call the predicate directly (the predicates in the included file are compiled as local object predicates). That just gives a performance penalty for no benefits. It also makes the comparison with the plain Prolog version an apples and oranges comparison.