I wouldn't get too concerned about occasional high numbers, I have also observed intermittent spikes in DH% even on earlier versions.
I think it's best to leave it set on --avalon-auto and let cgminer do it's thing, after all the programmers have put an immense amount of work into this and I'm pretty sure they know what works best.
@ckolivas, are communication errors used in the DH% calc? Considering the issues with the avalon controller this would explain the intermittent peaks.
I have also noticed that it takes a fairly long time for the DH% to return to it's previous average, would it be possible to reset it quicker?
Sometimes hardware errors come in runs. Then cgminer resets the avalon. The hardware error% you can see is the all time average, but internally cgminer uses a rolling average to determine what to do with --avalon-auto. If you want to reset the values, change the remote api settings to listen to another pc in your lan and use the zero stats function from the api.