I generally run with as much gain in the first stage as possible, then raise the gain of the second stage, etc.
BUT, from what I understand, the developers recommend threshold values in the 80-120 mV range (much higher than that and you risk some signals maxing out/clipping in the A/D converters; much lower than that and you may start to run into quantization issues with very low bit resolution of the waveforms, as well as too little signal-to-noise).
If you watch the auto filters, the threshold is all over the place (including far outside the 80-120 mV range). Even so, I generally do try to keep my thresholds around there.
What about 32*2*40 with a threshold of 90 mV (assuming you want that relative gain to be around 2900)?
Where I have my system now, I've found that I can usually run 32 * 2 * 40 with thresholds from 80-120 mV. The noise level is usually less than +/- 20 mV peak (I'm working toward +/- 10 mV peak). If I try to run higher gain (not relative gain, absolute gain setting) with strikes anywhere near me, I start to see some signals on the scope page maxing out when the y-axis is set to 100% (I'm not sure what the actual A/D max is, but it still makes me a bit uncomfortable that I may send distorted waveforms). PLUS, we have enough coverage in the midsection of the country now that there's not a whole lot of point in me trying to participate in signal capture from the west coast.
When strikes get closer, I reduce the relative gain accordingly, of course.