You can always try a different initial flow distribution. See what RAS is heading towards while going through the optimization. If it looks like the optimization runs are converging, but just run out of iteration attempts before it could get there, then use the flow distribution of the last iteration attempt, and then rerun RAS. If the optimization runs are diverging, or going unstable (i.e. solution errors are oscillating out of control, rather than minimizing for each iteration), then you may just have a geometry that is not conducive to split flow optimization. It could be that one of the channels is so much smaller than the other, or the ground points on one of the splits are so high, that the channel wants to go dry. In that case, perhaps a pilot channel would help.
If nothing seems to work, you can always use a stepdown scheme hotstart to let RAS determine the distribution of flow. http://hecrasmodel.blogspot.com/2010/12/how-to-create-hotstart-file-in-hec-ras.html
good luck