Would be easier if you provided sample data.

The basic idea is to use or conditions, which match first an escaped sequence, secondly a directive and then a single character.

With a /g modifier within a while condition the match will start where the last one ended.

Group only on your directives.

update

2 proofs of concept:

The second one w/o while loop.

Please notice that you need to check for defined, since every escape or single character is an empty (since ungrouped) match.

  DB<120> $str='xxx@@xx\@xx@abc\@xxx@efg@\xxx'
 => "xxx\@\@xx\\\@xx\@abc\\\@xxx\@efg\@\\xxx"


  DB<121> print ( defined $1 ? "$1\t" : "") while ( $str =~ m/ (?: [\\
+\@]{2} | ( \@\w+ ) | . ) /xg )
@abc    @efg    

  DB<122> grep { defined } $str =~ m/ (?: [\\\@]{2} | ( \@\w+ ) | . ) 
+/xg
  => ("\@abc", "\@efg")
[download]

update

more efficient:

  DB<135> grep {defined} $str =~ m/ (?: [\\\@]{2}+ | ( [\\\@]\w+ ) | [
+^\\\@]+ ) /xg 
 => ("\@abc", "\@efg")
[download]