Why can't I grep this way?

The spaces are more commonly known as "whitespace", and can include not just spaces but tabs (and other "blank" characters). In a regular expression you can often refer to these either with [[:space:]] or \s (depending on the RE engine) which includes both horizontal (space, tab and some unicode spacing characters of various width if available) for which you can also use [[:blank:]] and sometimes \h and vertical spacing characters (like line feed, form feed, vertical tab or carriage return). [[:space:]] is sometimes used in place of [[:blank:]] for its covering of the spurious carriage return character in Microsoft text files.

You cannot replace with grep - it's just a searching tool. Instead, to replace the yes with no you can use a command like this:

sed '/disable\>/s/\<yes\>/no/' /etc/xinetd.d/tftp

This tells sed to substitute (change) the word yes into no on any line that contains the word disable. (The \> (initially a ex/vi regexp operator), in some sed implementations, forces an end-of-word (though beware it's not whitespace-delimited-words, it would also match on disable-option)). Conveniently this sidesteps the issue of whitespace altogether.

---

Be careful: with a line such as eyes yes, an unbounded yes substitution would apply to the first instance of yes and leave you with eno yes. That's why I have used \<yes\> instead of just yes.

Why and how can I know what are the elements between "disable" and "=" ? Is it several spaces? tabulations?

Probably tabs. You could use something like

< /etc/xinetd.d/tftp grep disable | od -c

to see. It will show tabs as \t, and spaces as spaces.