Despite its potential as a reasoning paradigm in AI applications, abduction has been on the back burner in logic programming, as abduction can be too difficult to implement, and costly to perform, in particular if abductive solutions are not tabled. If they become tabled, then abductive solutions can be reused, even from one abductive context to another. On the other hand, current Prolog systems, with their tabling mechanisms, are mature enough to facilitate the introduction of tabling abductive solutions (tabled abduction) into them. The concept of tabled abduction has been realized recently in an abductive logic programming system TABDUAL. Besides tabling abductive solutions, TABDUAL also relies on the dual transformation. In this paper, we emphasize two TABDUAL improvements: (1) the dual transformation by need, and (2) a new construct for accessing ongoing abductive solutions, that permits modular mixes between abductive and non-abductive program parts. We apply subsequently these improvements on two distinct problems, and evaluate the performance and the scalability of TABDUAL on several benchmarks on the basis of these problems, by examining four TABDUAL variants.