Opening and closing a file is "expensive". It takes a lot of time.
And switching between files as well.
I've read something like wav and csv, 2 analog inputs. If both inputs read the same signal, then it doesn't make sense using two.
Since switching between files is also slow, you may use the same technique as WAV files for stereo audio: intercalating (alternating) samples. That is, log the first input, then the second, then the first again, then the second again, and so on; all into a single file.
Futhermore, because writting binary data is faster than writting text, you'll maybe have to generate the CSV file in a "post-processing" stage where timing constrains no longer exist.