@PhilProcarioJr: All good points, I agree. I was just trying to explain how this all works as it is today, not to try to imply that it is not confusing.
The XZ plane has always been a problem child, because of its odd axis orientations. The reason for the "upside down and backwards" part is because it was considered important that the sketch normal of a sketch on the XZ plane point in the +Y direction, so that a positive Extrude would go in the direction of the Y axis:

It helps to look at the coordinate system for a sketch on each origin plane. I've modeled a coordinate system using solid bodies, which makes it easy to rotate into various orientations. Here is the global CS (which is also the CS for a sketch on the XY plane):

In all these images, red is X, green is Y, and blue is Z (RGB...). This one is easy to understand.
Now, let's look at the YZ plane:

Remember, we are looking at the local sketch coordinate system. So X and Y are now rotated into the YZ plane, and Z is oriented toward the "inside" of the model. But, this is all kosher as far as text goes, which we have seen.
Now, the problem child: XZ. In order to get the default normal for the sketch to point along the global Y axis (toward the inside of the model), and to maintain a right-handed coordinate system, the local coordinate system of the XZ plane has to be oriented like this:

Which you can see if you rotate the axis model bodies.
So, it all comes from wanting that normal direction to be in a certain orientation. We could certainly change the normal direction for the XZ plane. Or make it an option. But, this is the way it is today.
I'm pretty sure nothing has changed in this area for a very long time, though. The "zero offset plane" approach never should have changed this, I believe, because an offset workplane inherits its coordinate system from what it was offset. So a zero offset plane from the XZ plane should produce the same results as the XZ plane itself. This seems to be the case, as far as I can tell.
I'm confused by the comments about drawings. Drawing views appear to be doing exactly the right thing: The view you get is exactly the same as the view when looking at the front view. Yes, the text is upside down and backward, but the drawing agrees with the model, doesn't it?
Thanks for the patience to have read all the way through this. I've also attached this model to this post, in case you want to take a look.
Jeff
Jeff Strater
Engineering Director