If you want to use BLASTN, it would be best to do so locally so that you can employ your own scoring matrix. If you cannot find a scoring matrix which allows G:U or G:T pairing, you'll have to build it. You'll want to change M and N, where the M parameter sets the reward score for a pair of matching residues and N sets the penalty score for mismatching residues. Typically M=5 and N=-4. If a G:T pair is weaker than G:C, maybe it should have a score of 1, midway between M of 5 and N of -4.

For more on scoring matrices, see this.

I don't know if it fast enough for you, but locARNA allows this kind of matching.