That was early research that wasn't incorporated into the software.
Unless you use a BYO section, ACBC will not oversample the most preferred levels within attributes (included in the BYO section).
One possibility that the software offers is the ability to ask respondents to specify a full or a partial ranking of levels for attributes in a question outside the ACBC question. Then, during ACBC/HB utility estimation, the utilities can be constrained (custom per person) based on responses to the "most likely" questions. This is described in the paper you reference...though the results were not as good as we had hoped for with that data set. Perhaps other data sets would show better results.
If interested, search the ACBC documentation for "Customized Constraints".