The first constaints on the boundaries of a 3.5mm window comes from the atmosphere: below 70GHz the attenuation of the atmosphere (even with only 4mm PWV) increases rapidly with lower frequencies. Observations below 70GHz will be increasingly difficult with decreasing frequency. The atmosphere becomes again the limiting factor for frequencies above 110GHz.
A second constraint for the upper limit of the band is the accuracy of the surface of the dishes. The efficiency of Mopra decreases steeply after 95 GHz currently. No other numbers are available at the time of writing.
The higher end of the end is determined by what is technically feasible; the higher the better, as the density of transitions increases rapidly.
In conclusion, based on the above and assuming that it is technically feasible to cover the full band from 70GHz to 95GHz, I propose (and define for the remainder of the discussion) that is the bandpass for the 3.5mm window.
For the requirements on the correlator at Mopra:
For most galactic work a bandwidth of few hundred km/s will suffice, with minimum velocity resolution of a few tens of km/s. This is currently avalable.
For surveys of transitions from objects in the local group of galaxies a bandwidth of a few hunderd km/s, with a velocity resolution of a few km/s will be sufficient. The desirable combination can only just be achieved with the present correlator capacity.
For large redshift surveys, the maximum bandwidth of 256 MHz would translate (at z=1.8) in a red shift window of only \delta z=0.018. Although the redshifts of low redshift objects (z=1-2) are known to high enough accuracy to target them with such limited window, this window is far too narrow to target high redshift objects. Furthermore, this window will make it impractical to search for absorption and emission lines towards low and high red shift objects.
Analogous to the situation at Mopra, the correlator capacity at ATCA is sufficient for galactic objects and objects in the local group of galaxies.
However, likewise to Mopra, the current correlator at ATCA excludes several scientific projects associated with high red shift objects from being practical.
In conclusion, the current correlator capacities of Mopra and ATCA are sufficient for nearly all kinds of research towards targest in the Galaxy and the Local group of galaxies.
To enable searches for medium (1-3) and high (3-5) redshifted lines is will be nec. to investigate the possibility to build filters and LO systems which would have bandwidths comparable with the size of the bandpass of the receiver: a few GHz.