back to quick_and_dirty, which is still practically important since the scheduler does not jump over subproofs;

--- a/src/Pure/System/isabelle_process.ML	Sat Nov 13 11:41:02 2010 +0100
+++ b/src/Pure/System/isabelle_process.ML	Sat Nov 13 12:32:21 2010 +0100
@@ -168,7 +168,7 @@
     val _ = OS.Process.sleep (seconds 0.5);  (*yield to raw ML toplevel*)
     val _ = Output.raw_stdout Symbol.STX;
 
-    val _ = quick_and_dirty := false;
+    val _ = quick_and_dirty := true;
     val _ = Goal.parallel_proofs := 0;
     val _ = Context.set_thread_data NONE;
     val _ = Unsynchronized.change print_mode