This doesn't really answer your main question, which seems to only be possible through some gymnastics. Maybe just cry uncle and don't attempt to combine the operation of validation and extraction:

  /^[A-Z]\d\d +1?\d +\d?\d(?:\.1)? +[0-3](?: +\d?\d\.\d)+ *$/ or ($bad
+{$line++} = $_) && next;
  my ($t, $p, $s, $d, @c) = split / +/;
[download]

And with use v5.22 you can clean the regex up even more by making all groups non-capturing with the "n" modifier:

  /^[A-Z]\d\d +1?\d +\d?\d(\.1)? +[0-3]( +\d?\d\.\d)+ *$/n
[download]

There is something to be said for keeping it simple and clear. So maybe even remove repeated space beforehand:

use v5.10;
use warnings;
my $valid = qr/^[A-Z]\d\d 1?\d \d?\d(\.1)? [0-3]( \d?\d\.\d)+ ?$/;
$, = ', ';

while (<DATA>) {
    chomp; tr/ / /s; say /$valid/ ? split : "Error:$_";
};

__DATA__
C31 6  3 2.4 1.5 2.6
C32 2  7 3 1.0
H31 1 1 0 21.0 11.2 5.3 1.4
T11 2  1 0 6.0 1.1 2.2
L06 1  1 0 1.0 3.3
L06 1  4 0 1.1 1.8
[download]