I do doubt that it would make engine any stronger but please do try if you like. I once tried something like that but did not help the engine at all. The GG evaluation is not really suoted to tree search. It gives values to moves not to boards situations. And good evaluation function for go are hard to come by.
Basically all good programs now are Monte Carlo Tree Search programs. Solves the evaluation problem nicely.
So if someone is still interested on improving gnugo the only thing to do would be trying out using gnugo knowledge to bias MCTS. You could experiment with pachi using GG to give a prior for search or something like that.