Its not non destructive. You're converting color spaces right? And you're altering the data so there's data loss. What would make Camera Raw handle the conversions any differently than Photoshop (it doesn't other than it has to do yet another conversion into a linear encoded ProPhoto space before it can do any further adjustments or conversions).
Correct me if I'm wrong, but I think the bit depth and colour space choices offered in the menu under the image in CR 4.x are essentially instructions for rendering the image. If one were to select say 8 bit sRGB and render the image on that basis, that is all one would end-up with in Photoshop; hence from a pure Photoshop perspective one has lost 7 bits of bit depth and the gamut between pro-Photo and sRGB. BUT, if one were to revert to the raw file and change those settings back to 16bit ProPhoto, is it not the case that all the raw data is still there allowing one to make that reversion?
Now, turning to the OPs issue, his problem starts with his description of the problem. Sure, if the image begins life in 16 bit ProPhoto he needs to get into 8 bit sRGB for the web. But Photoshop has excellent tools for doing this seemlessly. He should do all his image editing in 16 bit ProPhoto; then use Photoshop's tools to prepare his images for the web. All the stuff on my website was done as follows, starting from 16 bit ProPhoto:
Image Size with resamplingto 96 ppi and 600~800 pixels on the large dimension; interpolation bi-cubic sharper;
PK Output Sharpen for the web; (set opacity to taste);
Convert to sRGB with RelCol Intent and BPC on;
Convert Mode to 8 bit;
Save for Web in JPEG format, 60% quality, Optimized, with embedded sRGB profile.
That menu almost always works. Where gamut clipping still appears to be problematic, it is necessary to experiment with Perceptual Intent, and/or reduce the saturation of the offending colour group.
No other gymnastics in the raw converter are really needed.