I had a very similar problem on my DSO-X 3024A, in which the behavior of one channel changed when the attenuator relay changed state. I retouched the solder joints on L1 and R1 (see attached image, which is borrowed from the main hack thread). That restored proper operation. In my case, those components had been changed as part of the bandwidth upgrade mods that I had done earlier, so it was not the fault of the Agilent factory. Sounds like your situation is different, but perhaps a close inspection of that area is in order.
Good luck.