The next version has solved all of these problems (currently in the ramreduceleveldb and rrld_planB branches). It's not quite ready for testing, but should be very soon. Armory now builds the database and scans it on the first load only, saving everything between loads. And only uses like 250 MB of RAM (which should be independent of the blockchain size).
Obviously, this is a huge step up from where it currently is. I just haven't had the time to overhaul the relevant code until now.
Didn't you say that in March?
If its true, all I can say is...THANK FARK FOR THAT!