For everyone, I second the importance of maintaining the social life, relationships, fitness, and the like. Getting into regular exercise made a huge difference to my levels of focus, and participation in the local techie scene was instrumental to my own (ongoing) escape from such grunt work.
For internal career prediction, that's harder. My advice: just talk to people. The reputation of the project is a better indicator than any technical details. A project that is technically excellent but poorly sold internally is still going to be an effective failure.
I would evaluate external career boosts the same way I evaluate startup ideas. (Evan Williams has a good framework here:)
http://evhead.com/2007/12/how-to-evaluate-new-product-idea.a...
For internal career boosts, I would do the same, but eliminate the question about discoverability and monetizability, and view "wideness" and "deepness" through the lense of the power of individual executives within the company. It's not enough to hold a wide appeal to potential users, you need to hold wide appeal to the users that your VP cares about pleasing, or you need to deliver value along the lines of your VP's thought process.
I spent a lot of time investigating this in a number of different organizations. What I found is that most people who think they're overworked and have "no time" are spending a lot of time working on the wrong things. They've volunteered for mediocre projects and created too many recurring commitments for themselves. Recurring commitments make it harder to fire you, but they don't do much for your career, because doing the same job 20 times doesn't deliver much more benefit than doing it once.
If you work at a company where everyone works 9 to 8, you also have to work 9 to 8. The good news is that you can still probably get most of your work done in 2 hours, freeing up 9 to network internally and look for better projects, or read books/papers to study up on the skills your next job will require. If you never volunteer (except for a career-advancing project) and avoid recurring commitments, this means you have a lot of time to "learn on the clock". (Don't be dumb and write side-project code at your day job; you can do your exploration/research at your day job to make the side-project work more efficient at home, but it has to be done on your own time and machinery.) This is how most of the programmers who are reinventing themselves as "data scientists" are doing it.
Doing more grunt work, here meaning work that you don't learn or gain anything from, is a pointless game (for the Clueless). You get tired faster, you're more likely to lose social polish and fail politically. The first people fired or PIP'd in a "low performer initiative" (which is a dishonest layoff) are usually overperformers who worked too hard, got themselves emotionally overinvolved, and lost the political game. It's only in a deep cut (10 to 20%) that the company runs out of politically-failed "pain in the ass" employees and start removing the long-term underperformers who've been hiding for years.
Of course, there is work (growth-oriented work, in particular) where the returns are exponential and, for that kind of work, you shouldn't stop because it's 5:00pm. If one programmer is growing at 4% per year and the other one works three times as hard and grows at 12% per year, the difference over 20 years is that the former will be just 2.2 times more productive, and the latter will be 9.6 times more productive. (In capability, if not salary, I've seen programmers grow at 30 percent per year.) That may not matter for salarymen, but it's a big deal if you ever want to go off and start your own thing. The 4%-growth engineer is barely able to justify a senior-level salary; the 12%-growth engineer is a sought-after "10x" engineer.
So what's the answer? If the work is strategically valuable and growth-oriented, do as much of it as you physically can do well. This doesn't mean neglecting your social life or relationships or fitness, because those are important too. It does mean that you shouldn't stop just because it's 5:00 and you're three episodes behind on the Walking Dead. (I'd say that the average person can handle about 65 hours per week; beyond that, it starts to get unsustainable for most.)
For the scut work that's not going to help your career or teach you anything, that just gets shuffled around because no one wants to do it, do as little as possible. The politically optimal point, when assigned bad work, seems to be the 35th percentile in quality. Below the 25th percentile, you risk the appearance of having intentionally blown off the grunt work (which is different from just not being well-suited to it, because the former will be taken against you and the latter will be tacitly accepted) but above the 50th, you're in danger of doing it too well and getting assigned more of it.
The importance is knowing the difference. You need to do just enough of the "dues paying" work to fit in, and avoid recurring commitments as much as you can get away with, but put as much of your work time as possible into the compounding-interest growth-oriented work, and as little into the dead-end grunt work. In Iron Triangle terms, this is why strategy wins out over dedication.