I discovered that I am losing around 2-4 Mhps by using the WORKSIZE=128.  I stopped that.  I am convinced that dropping the mem clock does increase the hash rate for some reason.  I have tried it on two machines and am getting a better rate 2-3 mhps by dropping from 1000 Mhz to 500 Mhz.  I don't know why.  I am using MSI Afterburner to set clocks and apparently it won't let me go below 500, I may not want to for usability as others have said.  I am using phoenix not poclbm so my options are a bit different from yours.