Old thread, but my understanding is that you just feed all games into the glicko algorithm to get the overall rank. The result may be different than a weighted average of the sub-ranks. It may have changed, but I haven’t heard about that.
In psuedocode, the difference is this (only small and large boards for simplicity):
// in this example, [a,b,c] are game results from
// a "small" board, and [d, e, f] are game results
// from a "large" board
average = (glicko([a, b, c]) + glicko([d, e, f])) / 2
overall = glicko([a, b, c, d, e, f])