Our experience at Viewpoints Research Institute (a non-profit 501(c)(3) org) is that there are at least two very distinct processes with regard to "open source" (and these are often phased in time) . The first is to make something interesting in its own right that can also serve as a kernel. The second is to empower a community of interest...
We did Squeak, EToys, and Croquet pretty much the same way. First, we used our small research and development team of "extreme talent, knowledge and skill" to design and build and test the kernels. Because design at this level is not a simple matter of deciding what to do, but is much more error-prone and requires quite a bit of fast iteration and close communication, it is not possible or desirable to have a large open source community of interest kibitzing during this phase (even though a small percentage of the advice and help would be useful). Basically, there is too much noise of various kinds to permit real progress to be made. And we've found in our group that we can't really do design within our own process over the Internet, we have to get together physically for long periods of time to be effective.
It takes a few years to make a deep kernel.
We have put artifacts out on the net for people to play with early on, and this has sometimes led to misunderstandings and cloggings. The cloggings come when people want to develop on a kernel that is still unfinished. They want to do new stuff but get very upset if the kernel developers also want to do new stuff. Social pressure often leads to premature poor to bad defacto standards that impede progress (the current web and PC world -- and Java, etc. -- is almost completely at a conservative incremental standstill because of this).
Basically, my observation is that Sugar development is "clogged" (in case it is not obvious, Dr. Kay's comments were not in reference to OLPC).