Are exercises useful? Yes. But doing a significant fraction of them increases your read time substantially. It pays off, but now there are other ways (at work, or in open source) to learn things that are less structured but still effective.
Exercises help you get a deep knowledge of the material, but the market doesn't reward that. You're better off with a more shallow knowledge of a lot of things.
The old cycle for learning X:
1. Get an X book. Do a basic reading to get the vocabulary and core concepts.
2. Decide if you're interested in deep knowledge of X. If yes, continue. Otherwise, terminate and choose a new X.
3. Do the exercises in the X book. This will take a while.
4. Do some X at your job.
5. Read the research papers if you want to get to the frontier.
New process:
1. Get an X book. Do a basic reading to get the vocabulary and core concepts. (Same as above.)
2. Decide if you're interested in deep knowledge of X. If yes, continue. Otherwise, terminate and choose a new X. (Same as above.)
3. Get a job doing X. The knowledge and vocabulary learned in (1) should be enough to pass an interview.
4. Refer back to the book and, when needed, read for deep detail to supplement your on-the-job learning.
5. Read the research papers if you want to get to the frontier. (Same as above.)
Step (4) of the new cycle is where the exercises would come in, but most people by that point are just trying to solve the problem at work.