So, looking at the Calibre source code,
this is what sends the periodic NOOP opcode.
The
documentation for that method indicates that the GUI thread periodically calls it.
The smart device driver sends a NOOP every five probes. The 'normal' time between NOOP opcodes appears to be 2-3 seconds, so each probe is probably around 500ms apart (I haven't actually checked the code to see what it is).
The issue MGlitch appears to be having is that there seems to be something preventing the probes from occurring at a 'normal' rate. If Calibre is idle, then the only thing I can think of that would cause this issue would be a misbehaving plugin.
Of course, this is all a bunch of speculation...