Jodi, I found that it was not a memory leak issue, but as Curtis said:gp/arcpy scratch name methods can be very slow when the workspace in which you are generating scratchnames gets full of many files
Creating a new workspace for OUTPUT every 30 or so feature classes greatly increased the speed of my script. You are using listrasters to create the list before you enter the loop. In your loop, you could just change the name of the output workspace after a certain number of iterations. Below is some sample code. I didn't test it, so may have some errors, but should get you started.Sebastian#create scratch workspace name
tempDIR = "F:\\Data\\Reg09"
tempWorkspacePrefix = "update"
tempWorkspaceSuffix = 0
count = 0;
for raster in rasterList:
if count%30 == 0: # this is modulus operator, so should be true the first iteration and every 30 iterations thereafter.
tempWorkspaceSuffix = tempWorkspaceSuffix + 1
tempWS = tempWorkspacePrefix + str(tempWorkspaceSuffix)
if not arcpy.Exists(tempDIR + os.sep + tempWS):
arcpy.CreateFolder_management(tempDIR, tempWS)
out = tempDIR + os.sep + tempWS + os.sep
name = os.path.basename(raster).strip(".tif") # removes .tif from the file name
arcpy.env.cellSize = "MINOF"
arcpy.gp.ZonalStatisticsAsTable_sa(region, "VALUE", raster, out+name+".dbf","DATA", "ALL")
print" Raster " + str(raster) + " processing completed at " + time.asctime( time.localtime(time.time()) ).
count = count + 1
