Part of the problem was that broken version of the cnaf command was being used. The thresholds are reading back correctly in the software now.
I set the E and dE thresholds to the -50mV and -100mV, respectively, that Vince mentioned. The E threshold had been -300mV, but I think I did that by accident.
I certainly see the slowness you mention. I'm not sure what is driving that. As far as I can tell it is not a network issue, the controller+cards is just really slow sometimes. I seem to recall noticing this in 2008 too, but I don't recall a solution (if there was one). Note that the code has to poll every card individually and there is a set and a verify call. If each one takes a second or two for each cnaf command to respond, that adds up pretty quickly... Kind of painful though.
The long delay must be kept in mind when this code is added to any CODA scripts. End-of-run script only or the DAQ will stall. I would recommend that this just be added to the shift checklist so it is verified once per shift.