To verify, you wish to take data for 100 microseconds on each of 32 sensors and display the data in "real time." The problem is the overhead of the initiate/acquire, not the actual acquisition time or theoretical data transfer time. Have you considered taking all your data in a single acquisition? I assume you are already multiplexing the signal to the 5102. This will require a bit of work in setting up the advance trigger for the multiplexer and the reference trigger for the 5102, but this is far cheaper than buying another acquisition board. Your data acquisition sequence would then be:
With this method, you only take the overhead hit once per scan instead of 32 times. You should be able to get single-digit Hz update rates. You can speed up the data acquisition by using both channels of the 5102 (unless you are already doing this for some sort of data reduction). If you are doing this in LabVIEW, pass your data into a queue in your acquistion loop and pull it out of the queue in a separate loop to process and display. This allows you to time multiplex the acquisition and display.