What is the optimum insert length for an RNAi construct in plants?

Attempting to design a virus-induced gene silencing (VIGS) construct in wheat. From the literature I've seen, 300-800 nt antisense constructs are standard, but I've also heard successful accounts of much shorter hairpin constructs only 40-60 nt in length. Please share your experience with either of these strategies, plus any relevant citations.