Rising to @Ankan-Zerob's challenge, this is my estimate of the maximum length which can be stored in each text type measured in words:
Type | Bytes | English words | Multi-byte words
-----------+---------------+---------------+-----------------
TINYTEXT | 255 | ±44 | ±23
TEXT | 65,535 | ±11,000 | ±5,900
MEDIUMTEXT | 16,777,215 | ±2,800,000 | ±1,500,000
LONGTEXT | 4,294,967,295 | ±740,000,000 | ±380,000,000
In English, 4.8 letters per word is probably a good average (eg norvig.com/mayzner.html), though word lengths will vary according to domain (e.g. spoken language vs. academic papers), so there's no point being too precise. English is mostly single-byte ASCII characters, with very occasional multi-byte characters, so close to one-byte-per-letter. An extra character has to be allowed for inter-word spaces, so I've rounded down from 5.8 bytes per word. Languages with lots of accents such as say Polish would store slightly fewer words, as would e.g. German with longer words.
Languages requiring multi-byte characters such as Greek, Arabic, Hebrew, Hindi, Thai, etc, etc typically require two bytes per character in UTF-8. Guessing wildly at 5 letters per word, I've rounded down from 11 bytes per word.
CJK scripts (Hanzi, Kanji, Hiragana, Katakana, etc) I know nothing of; I believe characters mostly require 3 bytes in UTF-8, and (with massive simplification) they might be considered to use around 2 characters per word, so they would be somewhere between the other two. (CJK scripts are likely to require less storage using UTF-16, depending).
This is of course ignoring storage overheads etc.