We had done this at a previous company I had started (with others). We were over 200 people at the time (~ 75 developers), and had created a situation where there were serious clashes between the "start-up guys" (who preferred to put ideas out there quickly, warts and all) and the "serious engineering guys" (who didn't like the idea of supporting those things, and preferred no warts to ever be shown to customers). [0]
Our solution was to create a new team, with a new VP to report to, who was aligned with their goals explicitly. The team's focus was to simply find interesting new product spaces, prototype the solutions quickly, and get 1 paying customer. The "start-up guys" were moved to this team, and they hired more like-minded developers. The normal R&D department would take up the prototype as a proof-of-concept once it was worked out that there would be enough customers to justify the longer development of the final product. While it resulted in the creation of the company's latest product, it definitely did not solve the problem it was designed to - but it did increase the understanding of the engineering team as to what would, and wouldn't work when they got to work on the problem. Which, is really what everyone needed (even if it wasn't what they wanted).
Our solution was to create a new team, with a new VP to report to, who was aligned with their goals explicitly. The team's focus was to simply find interesting new product spaces, prototype the solutions quickly, and get 1 paying customer. The "start-up guys" were moved to this team, and they hired more like-minded developers. The normal R&D department would take up the prototype as a proof-of-concept once it was worked out that there would be enough customers to justify the longer development of the final product. While it resulted in the creation of the company's latest product, it definitely did not solve the problem it was designed to - but it did increase the understanding of the engineering team as to what would, and wouldn't work when they got to work on the problem. Which, is really what everyone needed (even if it wasn't what they wanted).
[0] - this is an over-simplification, of course