I received an off-list email from someone at Cisco which, as I understand it, said that the problem with bad memory was actually a design or manufacturing problem causing a units made over a period of time to exhibit this fault, and is less likely to happen with 256MB than 512MB.
He says that if I can provide the serial, which gives the dates of manufacture, he can check if it's one of the bad batch. He said the 5505 is effectively an embedded PC, and a reload doesn't power cycle the chips like other Cisco models. I got a chance to reload the other unit in the pair, and it's fine; my predecessor had never noticed the reload problem, so perhaps the faulty unit was previously OK but as it has aged a latent fault has occurred. I'll be raising an official Cisco fault report and see what happens.
