I think this paper is a great contribution. While modern registration methods were well evaluated by Klein et al. (2009)
http://www.ncbi.nlm.nih.g...
there have been few evaluations of recently developed segmentation methods.
It appears to be the first evaluation of the segmentation performance of SPM8's New Segment (though see also Hu et al. (2011) http://dx.doi.org/10.4028... for a blob-based evaluation) as well as the first thorough comparison between this and Gaser's VBM8.
The paper's use of both BrainWeb20 and IBSR for evaluating accuracy, together with an equally thorough evaluation of reproducibility, is a key strength, particularly given the authors' finding of different behaviour on real and simulated data, which is an important result in its own right.
This should provide a good foundation for future development of segmentation algorithms to build upon.