Statistics are not usually collected on dictionary tables. One possible reason is the information stored there changes frequently (example: DBC.ALLRights) and may therefore give scope for poor performance when the statistics goes stale.
In the absence of statistics, the estimates are sampled randomly on a AMP and these are fairly good in most of the cases (for dictionary tables).
Unless you regularly query dictionary tables, collecting statistics on them serves no purpose. The system-initiated dictionary retrievals that occur during SQL compilation do not benefit from statistics because they don't pass through the optimizer.