Specifically see RE (tilly) 6 (bench): File reading efficiency and other surly remarks. Your speed claim can only be made for the specific setup you tested. If your code will need to run on multiple machines then the optimization is almost certainly wasted effort. If performance does not turn out to be a problem, it is likewise counterproductive to have sacrificed maintainability for this.

In short, the fact that this might be faster is very good to know for the times that you need to squeeze performance out on one specific platform. But don't apply such optimizations until you know that you need to, and don't apply this one until you have benchmarked it against your target setup.

A few general notes on optimization. Given the overhead of an interpreted language, shorter code is likely to be faster. With well modularized code you retain the ability to recognize algorithm improvements later - which is almost always a better win. Worrying about debuggability up front speeds development and gives more time to worry after the fact about performance. And readable code is easier to understand and optimize.

Which all boils down to, don't prematurely optimize. Aim for good solid code that you are able to modify after you have enough of your project up and running that you can identify where the bottlenecks really turned out to be.