> For example the word const is not use anywhere in the 2k version, while Math. is used about 40 times so it becomes essentially free when compressed.

Except that the word const does seem to be used in the uncompressed version, so we don't know what happens during compression in the Closure compiler, no? Or is the 2k version pre-Closure compiler different from the uncompressed version?

Either way the only way to know is to try because asymptotically free is all well and good, but without measurement we don't know the actual behavior in practice and we are just making assumptions about how well or how badly things compress and how the mangling process works out.