That is the other thing, a roller dyno is so far from a proper load cell dyno. You really need a proper load cell to hit all possible cells and hold the car in a cell long enough to get steady state data. The repeatability of a roller dyno is trash as well, tyre temps, pressures and just plain position of the wheel on the roller all affect it too much. Obviously better than street tuning, but if you are going to find the time to hire a dyno, at least hire a modern load cell one.
The oem calibrators I've spoken to said it takes almost a second within each load point for the engine to reach true steady state, they said you can see it in the data that if it isn't held there long enough the data you get is wrong.
Where street tuning (or logging really) shines is creating a scatter plot of rpm vs load. You can then see if all you have properly scaled the breakpoints of the tables to match where you spend the majority of time. If the scatter plot is all bunched up in one corner you can quickly see where you should be rescaling and changing the break points. We added a "reinterpolate all referenced tables" function in the pcmtec editor exactly for this function, it auto re calculates all the cells to match the new axis, and when an axis is shared with multiple tables it does the whole lot for you.