I think I see the problem: the part is vibrating because its fixtured on top of a keyboard
I think Rob has the right idea. A model with surfaces in various orientations like that requires different machining strategies to tackle different slopes. Parallel finishing passes are good for low (shallow slopes) and Z level waterline is good for steeper slopes and walls. How efficiently you can do this in VM, I don't know.
Another thing to consider is the accuracy of your machine. Older machines that can barely manage 200 or 300 ipm rapids tend to have significant servo lag. This has an almost 'backlash-like' effect when reversing directions on a slope, because one axis is a wee bit slower than the others, so this causes the tool to drag on opposite sides of the commanded position. This exaggerates the scallop effect above what the program actually should produce.
The workaround for this might be single direction machining when using a parallel finishing pass. I know, lots of rapids back to the one side, but you do what you gotta do.
Another thing to watch is the climb/conventional milling attitude of the tool. Conventional does not cut as clean as climb. If you are not actively thinking about what you are programming, it is entirely possible to be conventional milling on a sloped surface when it 'looks like climb' because the actual contact point on the ball mill may be on the wrong side of center. This all depends on which way you are stepping up or down the slope, relative to the tool travel and tool rotation.