Number two is don't get involved in trying to fix the problem if things go wrong
That is good advice indeed.
And how would you prefer to deliver your images? I'm thinking in the context of: how do you "proof" (litteraly or figuratively) that you delivered as promised? Remember, the images are all in the digital realm. You say that they look good or correct on your calibrated monitor, but, as we have seen in another lengthy discussion, some people take that to mean an eye-balled adjustment of a vaguely related profile with completely insufficient Photoshop colorpreferences to boot.
Would you deliver an Epson gamut printed image? How do you proof the epson is well calibrated (in a quick & easy, per image, production sensible matter)?
Would you (also) deliver an Epson proof print of a SWOP converted version of the file, just to show what is to be expected with a default conversion.
Remember also that this is the primary reason that conventional prepress simply ignored the profile, and converts to CMYK at the first opportunity and make a conventional proof. (Let's call it "output-referred" ad extremum). Obviously you can't "ignore" the source profile, but for conventional prepress that would preferably be display co-ordinates. Part of the "expertise" of conventional prepress was in the ability of the operators to "translate" the screen image to actual print.
Once in the CMYK realm, another part of the "expertise" of conventional prepress is in the ability of the operators to translate measured values to print. This used to be a discussion in the time of the first chromacom systems. Conventional scanner operators didn't have a preview available in the old days, so they had to learn to always interpret a scan by the corresponding CMYK numbers. When digital image processing systems became available which included a relatively decent preview (with matte screens mounted in front of a CRT to emulate print-on-paper), scanner operators increasingly began to use the instant gratification feedback, which was (thought to be) detrimental to their ability to make scans the conventional way.
I think part of the slow transition in prepress to CM workflows (apart from an inherent inability to accept change), is that in the hay days of CM, this conventional method continued to work as the source image would mostly be in AppleRGB or sRGB. If you dumped the profile in favor of your own display representation (usually also AppleRGB in those days). this just results in slight colorshifts which would also happen in a CMYK conversion anyway. (You'd be surprised how many nuances of saturated Red would always be color corrected to M=Y=100%).
A lot has changed since then, and delivering a source image in AdobeRGB or ProphotoRGB will kill any possibility of such a conversion.
Now I think we face two main issues:
1. CMYK conversions remain an expertise because of all the other variables mentioned, such as black-generation and dot-gain compensation which btw isn't simply limited to adjusting for dot-gain, but should also include the minimum dot and maximum dot limits which incidentally can not be described correctly in the ICC profile model as it currently is.
Therefore I would recommend that Photographers should leave this to the next step in the chain, be it a prepress house, or the printer.
2. This means however that a Photographer needs some kind of proof that the images delivered conform to the contract or agreements.
The question thus becomes: How do you know (proof) that what you deliver is correct? i don't mean just proofing that the image has a profile embedded, because that still says nothing about how it looks in reproduction, even if the image gamut fits the reproduction gamut. Only if you actually have this reproduction do you have proof that you deliver something reasonable. And even then you might have to proof that the reproduction has been colormanaged properly.