A multitude of convolutive blind source separation algorithms exist, a small number of which can deal with moving sources. The main assumption for moving source algorithms is that for a small amount of time the sources are approximately stationary and hence the mixing conditions are slowly varying. In reality, speech sources are likely to fall silent and hence the mixing conditions will jump to new values. This paper compares a number of blind source separation algorithms focusing on robustness to source and jammer movement. Acoustic models of a single reflector, a studio and a meeting room are used to generate the source mixtures. In addition, weight robustness is assessed using real world recordings from a studio.