Hi Chinmaye,
Let me know if I am wrong, but I think you are not taking into account the bandwidth of the bias network. When you are biasing the amplifier using a 1:64 ratio (quite a strange ratio, I wouldn't use a ratio higher than 1:16), you have gm_out / gm_in = 64.
The bandwidth of the bias network is limited by the pole at the node where you generate the bias voltage (drain of the diode connected transistor). In this way, the GBW of that circuit will be ~gm_in/C_L, so when you reduce the ratio to 1:2 (assuming the capacitance does not increase too much), you are increasing the bandwidth 32 times! That means that the bias voltage will be recovered from any coupled noise faster.
However, this is just an assumption, we should need some waveforms and you should explain the problem thoroughly so we can understand it properly. Please check how stable is the current source in the opamp using both biasing networks and provide some feedback.
Regards.