Spacing of artificial miRNA target sites for efficient knockdown - (Oct/02/2008 )

I am making a construct expressing two shRNA hairpins against the same gene but my design software gives me target sequences that are very close together or overlapping on the mRNA (my target gene has domains homologous to other proteins).
What distance would I need between two sequences to ensure that they don't interfere with each other for RISC binding etc? Is there an optimum distance between sequences for more efficient knockdown?
Any thoughts?

According to this paper, a 4 nt spacer should be sufficient. PMID: 17694064