Charles,
 Is your time series from RUC analyses or from a RUC forecast? I suspect the former. Forecast time series of PBL height should be fairly continuous (except for temporary cooling from a precip event), but analysis time series may not be so.
 With daytime assimilation of METAR temps, the METAR may be a little cooler than the previous 1h forecast. In this event, the temperature higher in the PBL indicated in the 1h forecast will not be changed even though the surface layer is cooled in the new analysis. The result is that a shallow PBL is temporarily diagnosed. This is a difficult issue for assimilation of surface observations: what can we infer about the PBL conditions based on the METAR ob, and more importantly, do we know with certainty the difference between the situation when a deep PBL is cooler than expected in the background forecast, and when it is just a shallow cool pool near the surface.
 The upcoming 20km RUC has some improved logic about inferring conditions in the PBL when surface temps are higher than expected, but takes the conservative approach regarding the previous question when the surface obs temp is cooler than predicted.
 Thanks for pointing out this situation and I hope this answer helps. If PBL height is the variable of interest for your work, I suggest using a sequence of short-range forecasts in which the model physics has forced more interconsistency.

Stan Benjamin