I know that you have written the Fuji Frontier and other commercial photo printers do not have sRGB as their color space, but merely convert the sRGB data to their own gamuts. The assumption of an output device with sRGB characteristics was merely hypothetical.
Unfortunately, you still have not answered my original question. If you are using ProphotoRGB with a moderately sized gamut image and printing to a relatively smaller gamut device using perceptual rendering with a table based profile, does the gamut compression occur for the entire ProphotoRGB gamut or only to the extent of bringing the out of gamut colors into gamut?
The latter instance would require a "smart algorithm" that actually looks at the contents of the file to determine the extent of compression required, whearas the former would simply squash down the whole ProphotoRGB gamut.
First, the Frontier only assumes sRGB as the source color space when converting to the destination color space. In any conversion, you have to define the source and the destination. Both are equally important.
A working space is a container for holding you image data and it has a fixed gamut. You can shoot a gray card in RAW and encode that data into two working spaces of differing gamuts. It might fully fit within sRGB and it will of course fit in ProPhoto with a heck of a lot of additional gamut around this data. So looking at the scene gamut before even deciding what working space to use for encoding is useful. If you can fit this gray into sRGB, using ProPhoto buys you nothing.
OK you have a very saturated scene gamut that required ProPhoto RGB and you need to convert it into sRGB to place on the web. In this case your "suck" with RelCol gamut compression due to the profile structure.
You have an image and you want instead to map it to a printer color space and it's smaller. The compression will take place at the boundaries of the image gamut as you should see in the screen dump I placed in my last post. Hopefully you can see that there are colors that are not at the boundaries of the larger triangle. Some may be, some may not be but the area where a color lies in the chromaticity diagram is where the mapping starts. Does that answer the question?