From its description in the Adobe help, the Exposure slider should perform a linear scaling of the luminance data, just as decreasing the exposure by 1.0 EV scales all the picture elements by a factor of 0.5.
This can be shown by experiment with Imatest and a Stouffer wedge. With the ACR defaults, the black point is 5, which rolls off the shadows. The default Contrast of +25, Brightness of +50, and the default point curve are applied with the exposure adjustment and complicate the resulting curve.
Here are the tone curves with ACR defaults and Exposure set to 0 and -1:
Here are the tone curves with ACR defaults except for Black = 0:
And finally, here are the tone curves with ACR linear settings (black = 0, contrast = 0, brightness = 0, point curve = linear):
This last set of curves demonstrates that the Exposure adjustment is indeed linear over the entire range, except for the deepest shadows where the data are not good.
Panopeeper knows quite a bit in this area and also believes the Exposure adjustment is linear. Anyone who states otherwise should back up his statement with data.
I think Thomas Knoll's suggestions are for routine images in general, but there is no one more qualified to give such advise. I agree that one must evaluate the individual image. The example you gave with a generally underexposed image with blown highlights represents a high contrast scene whose dynamic range is greater than that of the camera. One must decide which luminances to favor in the final image. As you note, the order in which you do the recovery and exposure adjustments does not matter, since all these adjustments are concatenated and performed in the preferred order by ACR. As Schewe demonstrates in the tutorial, a curve applied to the highlights can also be useful and can give more control than the sliders.
OK- I see from the diagrams we are talking log-linear and that I can intuitively relate to what I observe by adjusting photographs. Thanks.