Have a look at Robert Smith's experiments on the accuracy determination of humidity sensors (http://www.kandrsmith.org/RJS/Misc/calib_dht22.html). Internal heating falsifying the measurements, a broken sensor or a non stable sensor were quickly ruled out. It seems my DHT22 is just not well calibrated when it left the factory.
I've been measuring humidity in a closed jar at the speed of 1 reading every 6 seconds (minimise internal heating) over 12 to 24 hours, with three different solutions : pure water, saturated NaCl and saturated MgCl. They should be reading 100%, 75% and 33% of RH. I took the actual readings, did a tiny linear regression on it, and have now a formula for converting actual readings into a calibrated output.