I agree that the best user experience comes from using Mathematica on the desktop. Any graphics rendering, let alone Dynamic content, using the cloud severely compromises performance.
Using Mathematica (purchase the license rather than SaaS) is, as far as I can tell, less expensive in the long run as well.
The only case where I can see cloud computing as an advantage is in cases where you can make use of a kernel in the cloud that has access to the proper GPU and enough cores to make the computation significantly faster than on the desktop. The computations would have to be sufficiently complex and time-consuming that the communication latency would be acceptable. There has been talk about offering this as a service at some of the WTCs I have attended, but there has been nothing official. It's like any other add-on device or peripheral. For exmple,is it cheaper to pay to have someone else 3D-print an object rather than buying a 3-D printer? I can (or will be able to soon be able to) buy a Thunderbolt cage for putting any GPU card I wish into, for roughly $1000-1500 USD. The extra costs for doing the computations in the cloud would have to be competitive.
In this case, of course, the cloud computing is really just a (rented) computer peripheral. It is not at all like trying to run a Mathematica notebook in the cloud from a Web browser. It's nice to be able to do this occasionally, but I can't think of any realistic case where I would want to do more than I can get Wolfram|Alpha to do where I would not have my laptop with me.
It seems to me that all the cloud initiatives are really just gateway drugs to get users to purchase Mathematica on their desktops. This is not a bad thing, necessarily. Let a K-12 student try the free or low cost cloud version to see if there is any interest, but then get a proper version. It's just like a student learning a musical instrument. Start out on a
$400 flute (for example), and then go for the $10,000+ flute if there is sufficient interest. The price bump for full desktop Mathematica is much smaller, of course.
Wolfram|One seems to be a product without a purpose, unless there is a marketing need to rebrand Mathematica so it doesn't seem so 'mathy'. You can have Mathematica with the identical subscription model as Wolfram|One, plus you can purchase a Mathematica license, which you cannot do with Wolfram|One.
In any case, if anyone is going to invest the time to learn Wolfram Language, the best course of action would be to get a desktop version of Mathematica. That way, there are no compromises at all regarding performance and flexibility.