3DMark 11 gives weird performance on full vs. benchmark-only (M17x R4)

Hey all. Been having some weird (maybe?) issues trying to figure something out about 3DMark.

They claim if you run the "Full 3DMark 11 Experience", only the benchmark portion is used to compute the score. Makes sense. I do this, and I get a great overall score, great GPU score, and a decent Physics score.

Then I run Benchmark only, my GPU drops by 100 or more points, my Physics rises by 100 or more points, and overall drops by about 100.

It's very consistent- each individually only varies by about 5-20 points.

What's weirder, say I run Demo only (no benchmark). Then I run Benchmark only. The new Benchmark scores will actually be identical to those of the Full experience (100 higher on GPU, 100 lower on Physics). Doesn't matter how many times I then run it- it's as though the "Full" flag gets set and doesn't turn back off. I need to actually run Full again, at which point Benchmark only drops down to 100 lower on GPU, 100 higher on Physics.

What gives? And why should a big rise in GPU come at an almost equal cost in Physics?

Thanks!