I've done testing in my lab with a lab scale ( weigh sample, boil off water, re-weight ) and have found that anything over 2 percent water can cause problems. Excess water will combine with the Lye in a side reaction, which lowers the amount of catalyst required for maximum conversion, resulting in lower quality biodiesel. To much water, and you wind up with "glop-soap"
If you use an "open" style of processor you can see the reaction and can "tweak" it a little by the addition of extra catalyst thereby saving the batch.