If level is out of adjustment it will show in the cumulative amounts of out of balance observations. On a run where backsights are 100' and foresights 200' on run from end point to end point with 9 turns in between, you will be out of balance 1000'. If the level is out .04'/100' you will miss by 0.40'. If you return relatively the same way you will close correctly but still have a bad loop. The only reason not to use the same turn points, is because it promotes this result unless deliberately avoided.
You could run randomly unbalanced legs and end up cumulatively balanced. This would result in good elevation at your end point but bad elevations on all of your turn points. Even running randomly unbalanced legs and ending up cumulatively balanced on the return of the loop through the same turn points you would close perfectly but still miss all of the turn points in between.
Agreed. Compensating error is a b*tch...
This is also a good argument for pegging the level before each project. It only takes 10-15 minutes to do, and with today's digital levels it's easier than ever. Pick your test and follow the instructions, adjust as directed.