what's the reason behind it?

The modules will handle any complications: embedded separation characters, quoting rules, missing columns, character encoding, EOL issues, etc. split will simply split a string into a list which is fine for trivial, well-formed in-memory CSV datasets. To go beyond such datasets would be tortuous without using Text::CSV_XS et al.

It is the same reason why you would not be recommended to use regular expressions to parse any but the most trivial XML sets, for example.

Update: (edited for typo)