Something you might consider is using the HR to FR ratio on a per cell or per bsc basis. I am currently working on such a scenario at the cell level. What i did per ericsson recommendation is calculate the half-rate to full-rate penetration. Then add this percentage to the known value of available installed tch resources per cell. This will essentially adjust for potential HR usage. Figure your erlangs using this new value.
It would be much easier on a system with HR deployed in all cells but i do not have such a scenario so i went to the cell level so that i have unique figures for each accordingly.
I am not certain on the accuracy of this method because i have just started to use it. Only time will tell.