Your parser is really just a tokenizer. m/.../gc lends itself well to writting tokenizers, as seen in the following snippet:

while (my $line = <>)
{
   chomp($line);

   my @tokens;

   foreach ($line) {
      /\G ( [^{\s]+ )/xgc && do { push(@tokens, $1);  redo };
      /\G ( {[^}]*} )/xgc && do { push(@tokens, $1);  redo };

      /\G \s+/xgc;
      redo;
   }

   ...process split line...
}
[download]

Cases not specified by the OP behave as follows:

"xxx{yyy zzz}" is split into "xxx" and "{yyy zzz}".
"{xxx yyy}zzz" is split into "{xxx yyy}" and "zzz".