Add comments, use 1-4-0 instead of 3-1-3

src/one-body-sample.parset

@@ -30,11 +30,19 @@ mu = 1
 nu2 = 3
 
 [solver.tnnmg.main]
-nu1 = 3
-mu = 1
-nu2 = 3
+# The canonical choice is 3-1-3. Since the nonlinear step is by far
+# the costliest, this is a lot more costly than 1-1-1. The number of
+# additional iterations is negligible.
+# A vvv- or vvvv-cycle appears to be even faster.
+# And leaving off the postsmoothing is even better yet
+nu1 = 1
+mu = 4
+nu2 = 0
 
 [localsolver]
+# Observation: Increasing this value yields noticeable additional
+# computational effort. The difference in the correction is
+# negligible.
 steps = 1
 
 [bisection]