Something perhaps more helpful but so far unmentioned (and somewhat OS-specific) is that statically linked executables usually fork & exec (especially exec) much faster than dynamically linked ones. This difference is usually only like 50..150 us vs 500..3000 us but can multiply up over thousands of files.

This only matters on the first run of `rga`, of course. While the dispatched-to decoder is likely mostly out of one's linking control, this overhead can be saved for the dispatcher, at least. So, I would suggest `rga-preproc` should have a static linking option/suggestion, at least on Linux.

Of course, this overhead may also fall below the noise of PDF/ebook/etc. parsing, but maybe not the decompression of small files in some dark horse format. :-)