My graphs were based on the H.264-BR timeline of the benchmark test, with one extra 5-th track, encoded to 720P, following the settings Tom used earlier.

That accounts for the differences. My own results were based on just a single timeline with just a single 1080i video track with no effects applied, encoded to 480i MPEG-2 DVD.

I will try the benchmark timeline next.

Update on my timeline test:

With MPE enabled, MRQ and maximum bit depth are already "enabled". Turning both of them on in Queued mode, in my system, actually forces both of those functions to be done in software-only mode (and thus heavily taxes the CPU). I confirmed this by looking at the GPU utilization graph in GPU-Z: It barely touched 10 percent with both MRQ and maximum bit depth forced on in queued mode, while often jumping to nearly the same level as direct-export mode with both MRQ and maximum bit depth turned off. Those two settings have no effect whatsoever in direct export mode.

Re-running the test (with both MRQ and maximum bit depth turned off since those settings are now redundant with MPE GPU acceleration) now shows that the performance in both queued mode and direct export mode are about equal. (In contrast, forcing both MRQ and Maximum Bit Depth on increased the rendering times in queued mode to a total encoding time of nearly triple that of the direct export mode in my system.)

I re-ran the test with my own simple 6-minute AVCHD clip to MPEG-2 DVD. I discovered that it is still faster in direct export mode than in AME queued mode - but the difference is much narrower. The queued mode still took 70 percent longer than the direct export mode although even the queued mode encode still took less time than real-time to complete.

So, the moral of my tests is:

If you're running a system with a CUDA card with PPro CS5 set to the MPE GPU-accelerated mode (either with an officially supported card or via a hack) and a "K.I.S.S" disk setup rather than a "L.O.V.E" disk setup, don't force MRQ and Maximum Bit Depth on if you're going to encode using the AME queued mode. If you do, you may force software-only mode which slows down your encoding times needlessly and substantially. In addition, the direct export mode, although faster than the queued mode on projects with only one video track, may become bogged down with additional video tracks, especially in HD-to-SD transcodes.