I would like to know the optimum temp fluctuation that can be considered..Suppose if I am simulating my system at 300K, I want to know the minimum deviation of temp that can be considered.
I would say that it could depend on the size of your system and on what you want to do with it. As an example, if you strain your sample, the occurrence of defects may influence temperature fluctuations. Nevertheless, If you just want to equilibrate your sample at 300 K, you should only have few degrees (I would say 5-10 K) fluctuations (I experienced that with EAM and Buckingham potentials).
Temperature fluctuations during equilibration pretty much have to do with the thermostat employed. However, during production, temperature will fluctuate within 1/sqrt(N), with N being the number of atoms of the system.
I usually do equilibration at lower temperatures than 300 K, but as mentioned above thermostat thermodynamic conditions variability require cross-thermostat calibration.
I think something like 2% fluctuation in temperature during equilibration is not big of a deal, actually sometimes having a room for temperature sampling can slightly mimic serial annealing minimization equilibration. However, as mentioned above we need to be very careful with these fluctuations in production runs, especially when investigating near-transition dynamics.
I agree with Luis Cota: statistical physics says that all types of thermodynamical variables fluctuate \propto 1/\sqrt(N), N being the number of particles. One point I want to mention: very important is the kinetics of thermalization. Different degrees of freedom are thermalized with different rates. When calculate the fluctuations, be sure that your system is in the thermal (Gibbs) equilibrium. Also note, that obeying the Maxwell's velocities distribution is necessary but not sufficient condition for the equilibrium..