It is a really impressive and powerful tool. Watch the tutorial videos as well to get an impression of the operation, workflow and creative uses.
As far as the heretic in you, it all depends on what you want to do and how you want to do it. As you get used to working with channels from all working color modes, you will get in the habit of analyzing and studying your image before you start plowing away in ACR or your raw converter. Some folks like to produce a finished image all in the raw converter of choice and even soft proof and print from it. Others don't.
I personally like to extract from the raw conversion process the best possible set of RGB data to start with. This is rarely a "finished" image - instead it is a set of data that retains as much detail and color information as possible. Often, this kind of approach lends itself to batch conversion for images shot under similar lighting or settings, etc., because I am not concerned with local changes in detail frequency, etc. that may require individual attention to each conversion. All that gets handled later. Similarly, there is little or no sharpening or NR applied, as it may not really be needed in that stage of the image editing process.
I usually work in Photoshop to do most of my edits. If, again, I find that I am making edits in a similar manner after working through a couple of images in a batch, I will make an action to automate the process. This is to condition the image for finishing touches. I usually experiment on a couple of images to see what needs to be done, and then have at the whole lot. I also like to work quickly and develop multiple quick edits on the same image and then combine the best of each image into a final composite - sometimes the color from one, the contrast and tone from another, the local details from a third, etc. You can't do this in ACR/LR, even though you can generate multiple versions of a single raw file.
While ACR/LR local adjustment brushes permit you to paint in some basic local edits, channels are the really powerful way to dial in an adjustment - usually you are targeting certain ranges of tone and color in an image and that's usually something a channel can isolate cleanly and elegantly. ACR/LR does not give you explicit access to this kind of data, so you are usually "auto masking" or brushing in your local edits. Time consuming and usually not very accurate - for me at least.
So, my workflow depends heavily on Photoshop, and that's where I do my thing. It does not mean that tackling A-to-Z in your raw converter of choice isn't the "correct" way to do things, it simply is not my cup of tea.
You can push a 16bit TIFF raw conversion pretty dang far outside of the raw converter.
As far as using C, M and Y channels - again it all depends on what you are looking to do. Remember that the "density" of those channels represents ink coverage, not luminosity (as in RGB). So, a cyan sky will be heavily black in the sky, which may be what you need to reject the sky in your image mask. Since the CMYK channel is ink density (that is, lots of it is represented as really dark) and RGB channel is luminosity (lots of it is really light) the opposing colors from each mode should produce similar, but not identical channels. For example, cyan's opposite is red, so the cyan channel and the red channel should look similar; however, one may be more suitable than the other in terms of masking or an Apply Image move - this variation is something to consider when maybe the RGB channel isn't quite getting you what you need. Take a look at the opposing CMY channel - and, you can always invert. Mostly, the K channel gets the skeleton of the image pushed into it and is good for reinforcing detail, so it is often the channel that gets used the most from CMYK.
Cool thing about Channel Power Tools is you don't need to know all that to use the CMYK channels - just generate a preview of all of the channels and you can pick which one works best for the particular situation. However, it is a great exercise to study the image and visualize what each channel will look like and then generate a preview to see if what you envisioned is "correct" - it is a fun way to learn and study how to decompose an image in preparation for identifying its strengths and weaknesses and targeting those aspects you can exploit to make the image better.