I think that Ross is close to the answer.
There is insufficient current in the circuit to properly "wet" the contact.
However, ploughing through the manufacturers bravado and bluster in their datasheet, it infers that 2mA should be OK, if I decipher the data correctly.
How about asking the manufacturer?
They seem to offer a 5 year warranty, present their applications guys with the facts and ask why their switches are failing.
Reed contacts switch pretty slow (especially the release), and the "magnet on a valve" construction probably leads to very slow changes in the magentic field. If the switching speed is low and you have 31v across the contact you might get some arcing causing the degradation problem.
Maybe try changing it to 2mA contact switching but at a low voltage like 5v. And add a nice slow RCR filter across the contacts which should make arcing nonexistant.
Then you only have to worry about mechanical wear which at slow switching speeds should be very low.
I would never have switched 31v dry with no snubbing.