I understand that people at Google are allowed to work on whatever project they want, at any time. If an employee working on project X decides they want to switch and work on project Y instead, they just inform their manager and then make the switch.
If you assume that:
a) people are most productive doing something they're passionate about; and
b) the set of Google employees is smart and creative enough that what they want to work on is probably the right thing to work on;
then this is probably the optimum method of allocating developer resources.
Before I arrived I was told that it used to work somewhat this way way back when. But while I was there it did not -- for the simple reason that it doesn't scale and just isn't practical. Gmail can't take on a dozen new engineers every week, nor can it lose a handful of crucial engineers.
You're encouraged to move on to another project every 18 months or so, but I'm pretty sure that doesn't happen too often, either.
Thanks for the clarification. The ad hoc manner in which many projects roll out at google, it almost seems as though they're still following the previous method; but I suppose that it could also be a function of management deciding to jump when a 20% project bears fruit.
If you assume that:
a) people are most productive doing something they're passionate about; and
b) the set of Google employees is smart and creative enough that what they want to work on is probably the right thing to work on;
then this is probably the optimum method of allocating developer resources.