Daniel suggested that the mystery noise that comes and goes might be a parasitic interference in the fiber.

This morning, before anybody started any noisy activities, I went into the optics lab and gently tapped the fiber that is hanging from the ceiling panel of the clean room, from 8:13 AM (15:13:00 07/Sep/2012 UTC) for 30 seconds.

Attached is the short time spectra taken at various times before, during and after the tapping period, and sure enough, only the trace with tapping (blue) shows a very high noise in the right frequency band while everything else looks the same.

If you're interested in the data file, find the file name in the screen shot.

Note that "everything else" is still somewhat noisier than the black reference taken on 22/Aug.