Sudhir,

1) The IP3 is figure of merit extrapolated from the
     "linear" region of operation. Since the IP3 exists
     outside the "linear" region, it is calculated by
     extrapolation.

     By "linear" we are refering the region where the
     output power increases linearly as the input power
     increases.

     You might  want to make sure that you are at least
     >10dB below the 1dB compression to make improve
     quality of the extrapolation.

2) As long as the circuit is behaving linearly, the IP3
   point should the same. When the output starts to
   compress, the extrapolation is no longer valid.
   
    The other factor to be aware of is that the
    accuracy of the third harmonic calculation is the
    lower limit of the extrapolation accuracy.

    You should check that the slope of the third-order  
    intermodulation product 60dB/decade. If the slope
    is less than accuracy of the simulation should be
    improved.

     BTW, there should be an app-note on simulating
     mixers on Sourcelink that should explain IP3 with
     pictures. Also the SpectreRF theory manual may
     provide useful information.

                                                        Best Regards,

                                                          Sheldon