First of all, thanks for the feedback, it is the main thing that allows us to improve the app.
Many of the problems you mention are hardcoded and can only be fixed in a future version, but I will at least try provide few temporary workarounds and hints.
U3D:
one hint: middle mouse button also rotates the view, but uses another method (the "virtual trackball") and it is a bit less sensitive; if you like this method more than the default one, you may eventually reconfigure the view and swap functionality of the buttons
Rendering:
Setting camera is a big problem; we have no idea, what's the best way to do it using the input devices available (mouse, keyboard). There are actually four distinct ways how to set camera and none of them is "the one".
We usually use this method:
rotate the object using the the right view (3D editor), then transfer this setting to the document using the "Camera from 3D editor window" toolbar button
Switching to Floating mode forces the UP vector to be aligned with Y+ axis (as much as possible while keeping view direction). It is sometimes handy to briefly switch to this mode after setting the camera using the 3D editor window.
Invert-Y controls the orientation of Y axis when rotating the camera with mouse (when mouse is not visible).
The preview and camera window discrepancies:
This bug is caused by different front clipping planes in OpenGL (camera) and raytracing (preview) devices. We are aware of it but did not fix it, because it did not appear often enough. Oh well.
The workaround is not to zoom-in so much. There is no problem when the 3D object does not fill the entire window - during conversion to icon, the empty borders are automatically removed to maximize the size of the object (or if you just render to image, you may run the operation manually).