I disagree. Games are a great starting point. Creating your own world where the user can immerse themselves and make a variety of choices is way more fun, on a visceral level, than converting Fahrenheit to Celsius or using jQuery to put a few special effects on your website.
That fun is part of a reward feedback psychological loop which helps get you motivated, and the motivation helps you study harder and get past frustration. So the positive effect on the learning process may be large enough to justify such an "impractical" application.
> games have no practical use
Let's not forget that computer/video games are a large, visible and highly successful part of the entertainment industry that essentially didn't exist before the computer era. We could also say that art, music, movies and fiction "have no practical use" -- but a large segment of humanity still finds them compelling enough to spend enormous amounts of time, money and other resources creating and consuming, and these things are widely considered to be a core part of what makes us human.
> beginners need to focus on the pure code concept
Depends on the person. I would argue that many people learn better when they can apply their learning to a non-trivial problem. This is why introductory university science courses have labs. This is why classes at all levels have large projects in addition to the tiny problems on homework and exams. In fact, an inability to apply your learning to a non-trivial project implies that the learning process is incomplete.
> requires spending attention on entirely orthogonal concepts
Most projects in the modern world require interfacing multiple languages, libraries and API's. Any web app, for example, will likely have -- at minimum! -- parts in HTML, JS, CSS, database, backend language, and a template engine. That's six orthogonal components.
If you're going to be dealing with multiple orthogonal concepts in the "real world" of programming, the "academic world" of programming (whether a formal college curriculum or just self-learning) would be remiss if it didn't give you some preparation for that aspect through exposure to projects involving integrating different technologies.
Even single-language projects involving only core libraries often require putting together different API's (for example, many applications can be roughly divided into dealing with files, dealing with a user interface, and internal core functionality.) In fact, not being able to modularize a large problem into manageable pieces that mostly treat other pieces as black boxes is a recipe for failure of any and all attempts at large projects.
While you will have to learn multiple API's and libraries to do anything money-generating, I think it's not an efficient route when first learning programming...why not do a simple project that will be immediately useful to you as soon you complete it?
The reason I wrote a simple game is it was an exercise I thought of to do that seemed hard. I did it quickly, correctly, and in my opinion cleverly. It made me happy. For the first time, my technical co-founders also agreed I had written the code the way they would have written it.
If you're intending to be a tech startup, why don't you code something useful for yourself, like something that turns an excel spreadsheet into HTML tabular data.