I think the more interesting question is, what changed? Is it simply too much time to put at risk as they scaled? Did they determine that 5 days of focus on just one thing works a lot better?
From Larry's POV, 20% time was supposed to be an escape valve to find the next billion-dollar business. He knew that good ideas often look like bad ideas and that there was a good chance the next great idea wouldn't come from him (after all, crucial elements of Google's success like the ad auction system didn't come from him), so he built in a mechanism where smart engineers could go around his back and work on a prototype without him knowing, and then only take it to him once there was a demo worth looking at. This unfortunately didn't really scale as the company grew bigger, because with more employees, the chance that someone other than Larry would shoot down your idea before it got to him got high enough that basically no good ideas were getting through. (The last 20% project I'm aware of that became a true product area was Google Now, started in 2007 and graduated to a full-fledged product around 2011.)
Craig Silverstein viewed 20% time as a way to build social ties across product areas and avoid siloing and trust barriers. The idea is that if you spent 20% of your time working with a team that had nothing to do with your main project, this created a parallel network, separate from the reporting hierarchy, that allowed information to flow across different departments. This usage of 20% time was still alive and well when I left, even though Craig Silverstein had left a couple years earlier.
Many managers and employees viewed 20% time as a way to "try out" a new team: if you were thinking of transferring, you could do a 20% project for the new team and they'd vouch for the transfer. This mechanism was also still alive and well, for those employees savvy enough to take advantage of it, when I left.
It takes a fairly massive amount of time to build something from scratch, and if you chose what you're working on, you'll want to spend that time. Managers are expected to adjust expectations and not require more than 4 days of work if a person has a 20% project, but given that it's software and software never seems to hew to planned deadlines, this tends to stretch.
Actually, I suspect that a lot of the decline in 20% time is because Google hires more people now who want a stable 9-5, who have no particular ambition to try out a new idea on their own or build something from scratch.