Tested the voronoi shattering feature on a model I downloaded from TurboSquid.
The algorithm that determines whether an object should be fragmented is rather dumb still (for example; you can see a couple of relatively small fragments cause larger objects to break apart in this video), but the fragmenting itself seems to hold up pretty well.
number of frames : 300
initial number of objects : 40
final number of objects : 3667
raw simulation time : 3m 20s
raw fragmenting time : 1m 26s
total time (including some MXS overhead which I need to either optimize or let the plugin take care of instead) : 4m 59s
specs : dual core laptop (2.26 GHz), 32bit windows vista, 4GB ram.