Another possible strategy would be to wrap those low-level DBI calls in logic that specifically catches error #2000, and responds by sleeping for a short-but-random period of time and trying the request again a certain number of times before giving up.   This becomes a targeted response to a specific error condition, applied only when DBI informs you that the error in question just did occur.

  sub doit {
    my $dbh = shift;

    # IF ASYNC ERROR OCCURS, RETRY THIS MANY TIMES
    for my $retry (1..10) {   
     
      $dbh->execute();

      # LEAVE NOW IF THE ERROR (IF ANY) IS OTHER THAN "ASYNC"
      return $dbh->err unless $dbh->err == 2000;  

      # ASYNC ERROR ... TAKE A RANDOM SHORT NAP
      sleep(100*rand());      
    }

    # OH DEAR ...
    die "asynchronous I/O error";
  }
[download]

Something like that ...