You can think about it this way. When rendering to disk, each expression, in each layer is called for each frame of your project. So, depending on the number of instructions per expression, the speedup can be exponential.
Render time = (# instructions per expression) * (# layers) * (# frames)
Of course, this assumes each expression is the same number of instructions. You can find a rough upper bound by using your longest expression. (Once again, assuming your longest expression compiles into the slowest code, which may not always be the case)
Best,
Sumanth P.