Convert Var statements to take advantage of local array as well... Gets slight improvements on a few benchmarks -- e.g. 2.7% on MD5, 1.5% on iBench, and no change in cordic. The var-intensive computation portion of Mikelima's Maldelbrot test #4 (dropping the 60% of the execution time that's spend setting innerHTML within KJS) shows about 7.5% improvement, though, but that was the motivating example, really.