Rank normalization will loss some parametric information
of expressions. According to the simulation results of a reference (http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3660216/), quantile normalization seems to perform better than rank normalization. For reference only!
Well said. On the other hand, nonparametric normalization (the rank normalization) has the advantage that it is very, very robust to outliers and anomalies in array data. So if your data is really crappy with potentially contaminated samples and/or very strong batch effects (experiments conducted by different labs/teams), rank normalization may be a better fit than the quantile normalization.