nothing is wrong - that's perfectly fine! Now, let's say you have an image that fits into sRGB excpet for some slightly higher saturated cyan tones. Then you'll process and store it in, say, AdobeRGB. That's fine, too! Next image fits into sRGB excpet for some slightly higher saturated cyan tones and some slightly higher saturated red tones. Then you'll process and store it in, say, ECI-RGB. That's fine, too! The next image has overall a slightly higher saturation and so you'll process and store it in, say, Hasselblad-RGB. The next image in Rec2020. And the next image in ProPhoto-RGB. And the next image in ACES. That's all fine!
So you have to watch out for every single image that you process in what intermediate color space it may fit. Why?
I think you're exaggerating a bit
. Three workspaces would be entirely enough and so would be two (sRGB and BetaRGB would be my choices).
There is no reason to use this workflow if you don't want to. The reason I use it is because I aim my workflow to the destination. So I don't (in general) process my images in the same way for print and web. It also keeps me focused on the image gamut and where it is going to end up. If I go straight to a very large workspace, after a while I find that I get lazy and just press buttons and don't even bother to soft-proof. If I constantly keep the destination in mind then this doesn't happen, and I find that my images end up being better.
What makes this relatively easy is that so much can be done in the raw converter - and of course I will always keep my raw images with their develop settings. It is true that if I know that I need to do a lot of editing of the image in Photoshop + plug-ins that I will go for the larger gamut working space and then target the destination, simply to avoid having to do the work twice. However there are things that do (or may) need to be done twice anyway, for example denoise and output sharpen.
You have to remember that I use smart objects, so that I can open the image from LR to Photoshop as a smart object, do things like capture sharpen and some basic corrections that I couldn't do in LR (some adjustment layers say), and then convert to my destination. As this involves setting the profile in ACR I have the advantage of being able to choose any profile I want (which I cannot do in LR). So I can even convert to my printer profile if I wish (and of course in that case turn of color management in the print driver).
Alternatively I can use virtual copies in Lightroom if that makes more sense for a particular set of images as I've mentioned before.
Whatever you do, if you want to optimize your colors you will need to keep color management in mind constantly. If you always go to ProPhoto, say, then you should really have soft-proofing turned on all the time, pretty much, unless you are prepared to make corrections right at the very end of your workflow, prior to output to web or printer (which really is far too late IMO). And if you are targetting both web and print then you will need to constantly flip your soft-proofing from sRGB to your print profile. If that's how you want to work then that's totally fine ... there's nothing wrong with it and there's no reason why an image processed using your workflow would be any better or worse than one processed by my workflow. It's just a question of preference and what works best for you.
The other way around...: you make a great deal of adjustments to your image and you know that the image gamut will never exceed sRGB - why not still store it in the actual color space of your RAW-software (where it comes from in any case!) and save it as 1:1 copy of your "original" adjusted RAW-file? Keeping the source profile doesn't hurt :-) And you will store all your "originals" in the very same source-colorspace and first convert to something else by application.
I take it you mean the camera profile, in the case of Capture One or DxO, say? You can't do this if Lightroom is your raw processor as I'm sure you know. If the camera profile is matrix-based and is a good one ... well there would be no fundamental reason that I can think of. It wouldn't suit me because I work towards the destination but it would be fine for your workflow. If the camera profile is table-based then you would need to make sure that it's a very good profile based on a large swatch or you will likely get interpolation errors. Cheers,
Robert