Mixing runtime and device API Any downside?

The Programming Guide warns against using both the runtime and device API. I’ve played around with using both, and I haven’t found any downside. Anyone know of any problems that can develop? Extra resource use?

“These APIs are mutually exclusive: An application should use either one or the other.” --CUDA Programming Guide v1.1, Section 4.5, page 29.