I assumed the OP was just enumerating the possibilities ...

That's a good guess, and it's my guess too, but it's only a guess. Without reading too much between the lines (or the data records), I suspect it may have been stevieb's intention gently to make a point about ambiguity in the OP. (graff makes this point explicitly.)

... the one-line regex is not very "readable" no matter how you do it.

How about if you do it in a readable manner? I like the approach of factoring out regex sub-pattern elements. While more verbose, this eases maintenance, supports use of common elements for other data validation and extraction purposes, and is easily extended.

c:\@Work\Perl\monks>perl -wMstrict -le
"my $na        = qr{ N/A }xms;
 my $n         = qr{ \d+ }xms;
 my $n_or_na   = qr{ (?: $na | $n) \s+ }xms;
 my $test_name = qr{ [[:alpha:]] [_[:alpha:]]* }xms;
 ;;
 for my $str (
   'N/A 17497118 basic_mem_test 17036us FAIL 1',
   '17497118 N/A basic_mem_test 17036us FAIL 1',
   '17497118 17497118 basic_mem_test 17036us FAIL 1',
   'N/A N/A basic_mem_test 17036us FAIL 1',
   'foo 12345678 fake_mem_test 999us FAIL 1',
   '12345678 foo fake_mem_test 999us FAIL 1',
   'foo foo fake_mem_test 999us FAIL 1',
   ) {
   print qq{'$str'};
   if ($str =~ m{ \A \s* $n_or_na{2} ($test_name) }xms) {
     print qq{  match, grabbed '$1'};
     }
   else {
     print '  NO match';
     }
   }
"
'N/A 17497118 basic_mem_test 17036us FAIL 1'
  match, grabbed 'basic_mem_test'
'17497118 N/A basic_mem_test 17036us FAIL 1'
  match, grabbed 'basic_mem_test'
'17497118 17497118 basic_mem_test 17036us FAIL 1'
  match, grabbed 'basic_mem_test'
'N/A N/A basic_mem_test 17036us FAIL 1'
  match, grabbed 'basic_mem_test'
'foo 12345678 fake_mem_test 999us FAIL 1'
  NO match
'12345678 foo fake_mem_test 999us FAIL 1'
  NO match
'foo foo fake_mem_test 999us FAIL 1'
  NO match
[download]