dgallup,
Your concern is reasonable, however I would say this problem may happen even when actual runout error of simple regular cylinder is inspected. Picture a shaft comprised of two cylinders - one is datum feature, second is toleranced relative to the first with total runout. In most cases there will be a fillet between the two cylinders. So during the inspection there will be also a possibility that dial indicator will go beyond toleranced cylindrical surface, touch the fillet and in consequence the reading will be distorted.
Bxbzq's part does not have to be inspected by continuous rotation 360 degrees around datum axis. This can be done slice after slice. The key thing is just to make sure that the readings are not zeroed between the slices.
Frank,
If the functional requirement on G. Henzold's picture is to keep the shaft parallel to datum plane A, why not to apply parallelism tolerance?
If the functional requirement is to keep the shaft parallel to datum plane A and located from datum plane A, why not to apply position tolerance (with possible refinement of parallelism, if needed)?
Why would somebody want to check circular runout in this case?