There've been a lot of people complaining in the WCG boards about the more demanding hardware requirements of the ud agent, particularly for HPF2.  In order for the results of both agents to be compatible the basic code would have to be the same, you'd think.  The difference must lie in the execution as the size of the downloads is about the same.  It's also curious about this validation business that BOINC uses.  The ud agent accepts results as is and credits them immediately but Boinc has to compare your result to at least two of the same, and as many as ten, before it will validate the result and give credit.  This process can take as long as a week. Seems like one of these agents is traveling much further to get to the same destination.
Many people are getting invalid results for which they get no credit at all, and this is p*ssing them off as well.
Boinc is the only way to go if you have Linux or Unix systems and/or you like to work on multiple projects.  It also supports multiprocessor systems.  But for people like me, who work on single projects with standard Windows computers and don't care about point totals, you have to give the edge to the ud agent for its set and forget simplicity.

john,
Your observation has me curious now. Perhaps you could post the question at WCG and see if we can get an answer?

For me, BOINC has posed fewer issues. The install of BOINC was easy, and hooking up with projects just takes a few seconds. The only complexity issue comes with the preference setting capability, that, to my mind, is beneficial. You could simply leave your preferences on default settings, with no additional time spent. The BOINC manager also has lovely logging capabilities that let you know what the system is doing. You won't find any logging on the "UD" client. I like to know what's going on. Plus, the added benefit of knowing my computer's time won't get wasted if one project goes down, BOINC wins hands down.