So, unless you are shooting canaries (or more realistically, very unusually intense sunsets - see Figure 6), almost all of the tones you can print or are likely to be able to print in the short to mid term future, are nicely contained within AdobeRGB.
This is an interesting discussion. I don't claim to have any great understanding of color management and color theory. I have Photoshop books on my shelf, and many other books on diverse topics, which I have not found the time to wade through; just peeked at occasionally.
But there are a few apparent facts that leave an impression on my simple mind.
(1) 255,255,0 prints out yellower on premium lustre on my 7600 from the ProPhoto colour space than it does from ARGB.
(2) RGB yellow on the swatches pallette (in PS CS )translates to different combinations of yellow and cyan on the info palette, according to the chosen working space; namely, sRGB 87%Y+8% C; Colormatch 80%Y+6%C; ARGB 97%Y+9%C; ProPhoto 100%Y.
These values are all out-of-gamut, indicated by the exclamation mark,
but in relation to the SWOP standard, I assume. But there's a question here that perhaps someone would like to answer.
Now to the canary situation that Paul refers to. The implication is, if your photo does not have exceptionally bright and saturated yellows, then using ProPhoto as your working space will serve no purpose and might even have disadvantages.
Now that certainly makes sense, but have we missed something? Your image might not have (indeed is unlikely to have) a pure RGB yellow.
But there might well be other shades, some quite dark even, that require that 100% yellow, in conjunction with the other primaries, that only ProPhoto can deliver.
Another point I'd like to make is in relation to banding in 8 bit. There seems to be some concept that because Prophoto stretches the gamut, the 0-255 values will be wider spaced and we might see gaps.
In defense of this possible downside (not a particularly good defense I might add because I can't now locate the source of the information) my understanding is a typical image in 24 bit color does not contain anywhere near the potential 16.7 million shades of colour. We should remember that this is the pallette to choose from. Any image that was required to represent all 16.7M colors would be of necessity an image in excess of 50MB without any extrapolation. Futhermore, every pixel would have to be unique; ie. a different value of RGB.
In practice, a typical 24 bit image (from vague memory) contains just a few thousand discrete colors. I'm reluctant to put a figure on it, but I think we're talking about 30,000, plus or minus. (Or is that the number of genes in the human genome ).