OK, so an update on this problem.
Since last posting there has been no recurrence until Sunday morning when it happened again. The PVR had been set to record 3 programmes that morning, but at ~12:30pm I noticed it was still recording when the final programme had already ended at least an hour previously. The clock on the front was also showing 08:30. I took the PVR out of standby and pressed stop on the remote to stop the recording. The dialogue came up and I selected STOP, but the dialogue just remained there and the PVR then became unresponsive and continued recording. After ~10 minutes I unplugged it from the mains, gave it 10 seconds then plugged it back in again. About 80% of the earliest recordings had had their sizes set to 0, and all of the recording schedule was messed up - it looked like it had previously thought it was ~1 month in the future, and adjusted the recording schedule accordingly before the clock was corrected after being power-cycled. All rather bizarre as well as annoying.
Then, after re-formatting the hard drive and sorting out the recordings, it performed all of its recordings correctly except for one on Monday evening. The recording was set for 2 hours, and it did record 2 hours, but the recordings list thought the recording was only 1 hour long, and when playing it back the time elapsed was running at half speed.
So today I finally had a chance to run some tests on the hard drive. I used the Seagate SeaTools disk diagnostics to run the Short Drive Self Test and then the Long Generic test. Both passed without error. Crystal Disk shows no bad sectors, and I created an NTFS partition for the entire drive and ran h2testw which passed without error.
All in all I'm none the wiser. I don't recall the bizarre clock behaviour previously so I'm unsure if that is pertinent, but I'm fairly confident that the hard drive is fine so the cause is something else.
Any more thoughts?
PS I'm very sure the jumping problem is signal-related - I get few jumps on Ch4 recordings, but quite a few on Ch4+1 recordings.
EDIT Reading back the previous post, the common problem of the loss of the EPG/recording schedule is exactly what happened when the clock went haywire on Sunday morning.