I’ve done my share of shell tricks in the past and I love them, but I got very disappointed when I noticed that the performance sucks for anything more data intensive in the shell. At least in bash, which was the interpreter that I’ve used. I ended up porting my code to Python and gained at least 10x in performance.

But I prefer the shell code for tasks interfacing with system tools. I wonder if there is a shell interpreter around that is really fast.

https://github.com/oilshell/oil

Oil isn't fast at the moment, but I'm currently optimizing it (auto-translating some code to C++, and eventually hand-optimizing other parts.) I run benchmarks on every release, so speed is definitely a goal.

I haven't seen that many shell scripts that are "CPU bound" (as opposed to waiting for processes or I/O). (Although now I recall a shell-script-library bundler project that was.)

Of course, one big problem with shell is that you often have to spawn an external process like sed or awk to do string manipulation that's just a library call in Python. That could easily be a three order-of-magnitude difference -- e.g. milliseconds vs. microseconds.