Exactly. Plus the processing in LR is linear encoded, the processing outside (Photoshop) is gamma encoded. HALF of all the data in a linear encoded doc is in the first stop of highlight (read Michael's Expose to the Right). Once you distribute this in a gamma corrected, pixel based doc, lots of image processing options like control over highlight recovery are either very difficult or impossible. LR is using metadata instructions to build a big "edit list" of how it will render new pixels from the Raw data as you describe. Once you've got baked pixels in Photoshop, making big moves is again, either difficult or impossible and there's a huge speed hit (rendered pixels are slow to alter, metadata isn't). Unbaking a cake to alter the amount of sweetness is real difficult (pixel editing). Defining or altering the recipe prior to baking (altering the metadata instructions), is fast, easy and flexible.
The brightest f/stop of a 14 bit linear raw file does contain half the total levels in the capture, but not half of the usable information, since as
Emil Martinec explains in his excellent analysis, most of that precision is used to record shot noise (which increases with exposure). In this brightest f/stop, the shot noise is often 10 or more raw levels, and many of these superfluous levels can be thrown away without visibly affecting the image (see the section on lossy NEF compression). The real reason for exposing to the right has little to do with the number of levels, but rather the improved signal to noise ratio that results from the increased exposure.
If a gamma encoding is done with sufficient precision (16 bits / channel should be sufficient) with no clipping, the encoding can be undone by an inverse gamma curve and the resulting linear file can be processed as any other linear file. For example, contrary to previous assertions, a 16 bit TIFF can easily be white balanced in ACR. The process is complicated in practice, since most raw converters apply more than a simple gamma encoding.
