For implicit solvation with no friction, there's not much physical meaning for these thermal fluctuations. The magnitude is somewhat large because you have only a single independent particle, which probably means you'd have to run the simulation a lot longer for reasonable statistics in any event.
As an aside, the NOSE dynamics with the VVER integrator may not be a good choice for this. You'd be better off running Langevin dynamics, probably at low friction (SCALAR FBETA 2.0) if your main interest is conformational sampling. The Langevin bath gives you temperature control, and the low friction tends to give more dihedral transitions w/o affecting equlibrium distributions. Also, the VVER integrator is suspect, and has been replaced in newer CHARMM releases.