So, we have three characters different than a space and then a space: C, P, U and space, so from 0 to infinity no-space characters, the longest match is CPU:

Code:

% print 'CPU Util SSUUUUU | 16% 9% 22%'|sed -r 's/^([^ ]*).*/-->\1<--/'  
-->CPU<--

First of all, when the sed regular expressions engine tries to match that pattern, the string CPU is already consumed by the first match.
So it now is trying to match from the space that follows the string CPU onwards and, considering the string above, we have: a space, U, t, i, l, a space, S, S, U, U, U, U, U, one more space, pipe and another space and this string is matched by the following regular expression: [^0-9]*:

Code:

% print 'CPU Util SSUUUUU | 16% 9% 22%'|sed -r 's/^([^ ]*)( [^0-9]*).*/-->\1<-- -->\2<--/'
-->CPU<-- --> Util SSUUUUU | <--

And now the final part: [0-9]*%. The sed regular expressions engine is searching from the space after the pipe (excluding the space itself) onwards. It tries to match any number of digits followed by the % character, so it matches only 16%:

Code:

% print 'CPU Util SSUUUUU | 16% 9% 22%' |
  sed -r '
    s/^([^ ]*)( [^0-9]*)([0-9]*%).*$/-->\1<-- -->\2<-- -->\3<--/
'
-->CPU<-- --> Util SSUUUUU | <-- -->16%<--

If we had included a space and the % character in our character class, we would have matched the entire string,
leaving nothing for the last .*$ part:

Code:

% print 'CPU Util SSUUUUU | 16% 9% 22%' |
  sed -r '
    s/^([^ ]*)( [^0-9]*)([0-9 %]*%)(.*$)/-->\1<-- -->\2<-- -->\3<-- -->\4<--/
'                                                                           
-->CPU<-- --> Util SSUUUUU | <-- -->16% 9% 22%<-- --><--

Because the * mean 0 or more. If we force the last regular expression to match 1 or more[1] (not 0 or more) characters,
the result will be different:

Code:

% print 'CPU Util SSUUUUU | 16% 9% 22%' |
  sed -r '
    s/^([^ ]*)( [^0-9]*)([0-9 %]*%)(.+)$/-->\1<-- -->\2<-- -->\3<-- -->\4<--/
'                                                                           
-->CPU<-- --> Util SSUUUUU | <-- -->16% 9%<-- --> 22%<--

[1]. Only a few sed implementations support extended regular expressions (the + quantifier).