Idea

The blocks are broken up because the state of the processor is broken up. The processor wasn't processing for a solid 20 seconds. It was processing an item for 10 seconds, then immediately began processing a second item for another 10 seconds. Even in this simple case, merging the data would obscure the fact that the processor switched items. If the bar was simply 20 seconds long, you would not be able to tell if the processor was working on a single long job or on two short jobs. This seems like very important information.

Showing the separate bars also exposes how the 3D objects deal with state. They really do change state, even if it's for zero time. I'm not inclined to make it look like they don't change state as much as they do.

The gantt chart is also not just for drawing state. It draws whatever gantt-like data the user chooses to collect. I suspect that it is usually important that two same-colored bars on the same row are, in fact, two separate bars, and not a single bar, even if one starts exactly where the other ends.

So all that being said, you can always just collect "fuzzier" state data. You can use a process flow like the one in the attached model to ignore zero-time states. That way, the gantt chart isn't obscuring data; that data was simply never collected in the first place.

gantt-invisible-states.fsm