I'm saying that until proven, optimizing for "eliminating temporary variables" is likely to be a very pointless optimization. Much more interesting to optimize for minimizing the number of times the data is being copied, or number of opcodes executed, or total memory allocated, or needless actions repeated. "Temporary variables" can either be cheap or expensive, so it's like optimizing to remove the number of vowels in the variable names -- it's a red herring.