As for your the other points you raise about what to do when the marketing team is out of work: perhaps that would be an issue at other companies with less growth, but so far in about three years total working with Knewton (as a contractor and then as a full time employee) there has never been a moment where any of us were struggling to find something to do. There is a constant flow of large projects and firedrills and plenty of work to do in the icebox. These are projects that are important to the company's growth and vision, but would distract the product engineering teams from their jobs.
The product teams have their sprints planned out and aren't able to punt on an important feature because the biz dev team needs help building a mock application for a meeting next week. The marketing team has the ability to set our own priorities and timelines and adjust those as the business's needs change. We are better able to react to emergencies than larger teams because we aren't beholden to clients and product managers to meet timelines established months ago. I am not trying to criticize how the other teams work--I understand the benefits of that workflow, and I can't imagine how you'd build large, complex applications without that kind of organization. However, for the kind of projects the marketing team handles, that sort of organization would be more of a hinderance.
I'm on Core Engineering and we build the products. The Agency engineers figure out cool ways to use the products. Like they did a live celebrity selfies for the Oscars.
Are customers building things for the Oscars? Not really, but it shows the power of our products and helps marketing/sales. Which in turn let's core engineering experiment and contribute to open source projects. That leads to better products (and happy engineers.) Agency uses those new/better products to repeat the cycle all over again.
OK...without getting into "when is a web developer also an actual engineer?" debate...what do these engineers do when the marketers don't have any big ideas at the moment? Engineers on an engineering team ostensibly always have something to work and iterate on, directly...but if they're at the behest to the marketing team, then that's not a particularly agile process.
Well, I mean it could be, but then you've set up a shadow engineering team that is basically staked with creating whizbang ideas to compete with what the non-marketing-engineers are toiling to maintain. Actually, if I were in the non-marketing team, I'd find that quite annoying, as the old-cash-cow is always more annoying than what the bleeding-edge dev team is working on. But this is even worse than the Apple ][ team competing with the Macintosh team...the marketing-engineering team, in the OP's description, could really go down the road of creating plausible vaporware, sucking the attention of the company management.
The OP does give a good real-life example of what the marketing-engineers can do: a documentation portal for API developers, and visualization tools...two things that could be from the main engineering department but could also be completely separate...yet require key technical skills. But in that case, it's often just kind of another engineering team, at a mid-to-large-sized company that may have several teams (like how Google has search, local, social, whatever Peter Norvig works on, etc).
I guess I just don't see many situations in which building out such a parallel engineering team would be feasible for most startups, where the duplication of certain infrastructure and procedures is far more costly than it is for an established company.