How PBAT splits extended pedigrees can give disparate analysis results.

Getting disparate results in PBAT when running the exact same analysis is most likely the result of how PBAT handles extended pedigrees. The “Maximal Number of Non-Founders” parameter and the sorting of the spreadsheet can influence this.
 
PBAT splits up pedigrees if they have a larger number of non-founders than the maximal specified value that they should have. The sorting order does matter in how a family is split. There is not necessarily an optimal split since it starts with the first nuclear family and then splits accordingly after that. One way to get consistent results would be to sort by the proband ID every time, but again the results you get are going to be approximations. The only way to get a similar result every time would be to ensure that your maximum number of non-founders is at least as large as that of your largest pedigrees.

HOWEVER: To use a high number of non-founders can result in prohibitive computational time unless you select “Maximal Number of Non-Founders”. If you wish to perform power calculations, it might be useful to pre-scan your data, either using a smaller number of non-founders or using the non-parametric approach (the latter for good p-values), and then re-run the calculation only on the better candidate markers using the higher number of non-founders and “Screening Based on Conditional Power”.

The bottom line is that anytime you are splitting up pedigrees, your answer will be an approximation and there isn't necessarily an "optimal" split. The only way to get the consistent results is to make your maximum number of non-founders as large as your largest pedigree.