Do homologous arms need to overlap right at the DSB for knock-in - (Apr/17/2014 )

So I'm attempting this CRISPR/Cas9 system to knock in a gene at the Rosa26 locus. I will be using 750-1000bp homologous arms on each side.  My question is do the homologous arms need to overlap right at the double strand break or can I go up (5' arm) or down (3'arm) a little?

My understanding is that the efficiency is optimal when the homology arms extend all the way to the DSB but I know that this does not have to be the case in order to achieve HDR (it was achieved at our lab without such a perfect donor). I don't have any numbers to give you though, regarding how far you can go and I don't know how rapidly efficiency drops.