Make an existing workaround a bit broader.

Without this, the test seems to fail after clang roll
https://codereview.chromium.org/2577523002
Since there's already a workaround, this is likely triggering
the same existing bug that already has a workaround, and not
due to a new compiler bug.

BUG=673463,674194
TBR=tkent

Review-Url: https://codereview.chromium.org/2569763006
Cr-Commit-Position: refs/heads/master@{#438595}

third_party/WebKit/LayoutTests/fast/forms/associatedFormControls-leak-nodes.html

@@ -12,7 +12,7 @@ if (!window.internals) {
     window.jsTestIsAsync = true;
     var documentsBefore;
     var documentsAfter;
-    // FIXME(keishi): Calling asyncGC twice to fix flakiness.
+    // FIXME(keishi): Calling asyncGC twice to fix flakiness, crbug.com/674194
     asyncGC(function() {
         asyncGC(function() {
             documentsBefore = window.internals.numberOfLiveDocuments();
@@ -22,6 +22,8 @@ if (!window.internals) {
             document.body.removeChild(frame);
             frame = null;
 
+            // FIXME(keishi): crbug.com/674194
+            asyncGC(function() {
                 asyncGC(function() {
                     documentsAfter = window.internals.numberOfLiveDocuments();
 
@@ -31,6 +33,7 @@ if (!window.internals) {
                 });
             });
         });
+    });
 }
 </script>
 </body>