If you search the forums for "race condition" and "highlight execution", I'm sure you'll find numerous threads about how programs behave differently under highlight execution than under full speed. This behavior is an indication of a race condition.
You'll hear the term "dataflow" used a lot. Bascially, LabVIEW will execute a piece of code (such as a subVI, addition operator, whatever) when all of the inputs to it are available. If two pieces of code have no dependence of one versus the other, then whichever gets all of its inputs first will likely be scheduled as the next to run. At full speed, timing differences on the order of milliseconds or even microseconds between the two individual paths could cause one or the other piece of code to execute first. With highlight execution, you've essentially slowed the operation down by thousands of times, and LabVIEW is running extra code in the background to show you the execution process. This all affects how the entire program operates. This may cause the program to operate more in line with how you are hoping (as in this case) or the direct opposite of how you hope.
