As Dave said, this sort of thing does keep cropping up from time to time. It's even been useful; I remember people slicing the top off memory chips back in the day to use them as crude cameras.
More often, it's that kind of bug that keeps you baffled for a bit. In my case, I was writing the low-level code for a Z80 product's floppy drives (the Spectrum +3, in Amstrad's palatial offices), and the test system was a reasonably huge early wirewrapped prototype that took up most of the desk. The dev system was an Amstrad PCW8256 (we used CP/M, it worked well) with a homebrewed hard disk and a Z80 ICE, also fairly chunky, so I couldn't easily sit right over the floppies and run test code at the same time. To make things a bit easier, I blu-tacked (that stuff again!) tiny flags on stiff wire to the heads of the bare floppy drives, so I could sit at the keyboard and watch the head movements from afar.
It was all fairly simple stuff - the NEC uPD765 controller was a docile beast - and one of those satisfying little tasks that lets you feel you're making a lot of progress quite quickly (and more importantly, so does the boss). But every so often, one or the other, sometimes both, of the drives started to misbehave. The 765 status regs returned curious values, but if you did things like swap the drives the fault stayed with the logical, not the physical, drive. There was nothing up with the breadboard or the controller chip... had to be my code, right?
Pleasurable progress on hold. Often I'd give up at the end of the day and come back in the morning to find everything was fine again for a few hours, but as home time approached the problems came back. They'd go away again later, after I'd missed my train home.
It finally hit me that the fault occurred at roughly the same time every day. The culprit, unusually for Essex, was the sun. As it made its daily path across the sky, its rays tracked through the windows across the office - and when it hit just the right bit of the test setup, it shone on the optical sensor in the drive that spotted the floppy index hole by means of an IR LED shining through from the other side. The result was indeterminate but unwelcome.
Tape applied. Progress restored. Code exonerated. Gods blamed...
R