Good question Allison, and since I'm on-line I'll start a response, but a better answer would probably come from others on the site (such as scottsvb or Jason Kelley - hint, hint) that use the model outputs far more than I get a chance to on a day-to-day basis. As an old weather programmer, I know that we did have 'target' times that we tried to meet for each model run, and I'm sure that is still the case at NOAA. Sometimes a run would be placed 'on hold' if an entire block of data had not yet been received from a particular sector of the globe or group of reporting stations - usually because of some communications problems - so the initial analysis would not be as accurate and therefore neither would the forecast charts. Another reason for a late start is the computer itself - the schedule will easily get behind if the system has been 'down' for a period of time. I'm sure that I've missed a few other good reasons that others can help with to give you a better answer.
Cheers,
ED