Generally, one is interested in the differential signal to differential noise ratio, so common-mode noise is rejected somewhat. But in real circuits, especially in mismatched conditions when CM rejection is poor, common-mode noise can then become differentail and problematic. A monte-carlo sim will show you.

In you case, I would definately be looking at the Common-mode feedback scheme you are using in your OTA. I think you might find it improves if instead of using differential capacitance to get your TF, you use the capaciators in common-mode (at the cost of of 4* cap area)... this will should give you a noticable change in common-mode response.... and usually CMFB stability is the scurge of your type of design... i've seen a few high Q gmC filters oscillate nicely in common-mode...  

Anyway in cascaded biquad design, there is always an optimum cascade order for either noise or linearity/signal handling. Usually one choses the choice for best dynamic ratio (ie signal handling to noise ratio). So changing the order of the biquads will change the noise.
Also remember if the bandpass is real and created by bilinear transform of low-pass transfer function, then it will be symetrical in log(frq) rather than linear frq, so the noise always grows faster on the low side of the characteristic than the high side (ignoring effects like flicker noise etc.). This is not the case in your problem as you are quote noise at the -3dB point.... rather than just of of band... but its something to bare in mind.

cheers

aw

thanks,

yeah it was noise from the common mode detector. I was hoping i could get away with using a very simple common mode biasing scheme but it looks like i'm gonna have to do the whole CMFB loop.

Aaron