FORDISC will only return a correct ancestry attribution when an unidentified specimen is more or less complete and belongs to one of the populations represented in the program's reference samples.
1) Of course a specimen has to be "more or less complete" to be correctly classified. That applies in forensic anthropology in general and has nothing to do with the efficacy of the FORDISC program. No program or anthropologist can be expected to accurately classify highly incomplete crania, which is what the authors have simulated in one of their tests by reducing the number of cranial metric variables from 56 to 10. This argument is the equivalent of faulting someone for not being able to identify the image on a puzzle that's missing 80% of its pieces.
2) Under Materials and Methods, the authors explain that they only consider an ancestry attribution "correct" if the specimen is assigned to its source population or one that they've selected as its "most closely related population". So for example, they match up Berg (Austria) with Norse (Norway), and if a Berg specimen clusters instead with, say, Zalavar (Hungary) — Austria's next-door neighbor — or Gizeh (Egypt), they'd consider it "misclassified", even though the program has correctly identified it as West Eurasian/Caucasoid. Limiting the criteria for correct classifications to exact matches with source populations or, in the absence of these, pre-selected populations of questionable equivalence, is setting the experiment up to fail.
Marina Elliott and Mark Collard (2009). FORDISC and the determination of ancestry from cranial measurements. Biol Lett., vol. 5 no. 6, pp. 849-852.