MapReduce. Leveraging lower level abstractions incipient in the early Google datacenter (distributed FS, distributed process executor) and providing a simple but expressive programming model for distributed large scale batch computations. A significant amount of effort went into documenting and exemplifying the usage of the programming model for a large number of use-cases. Chosen to serve thousands of use-cases all across a fast growing company. Developed by 2 people. A fantastic example of '10[00]x engineering', and yet I can't imagine something more antithetic to the 'lone gunman' strawman.

I think Jeff + Sanjay is arguably a better example of 100x developers. Goes way beyond just MapReduce.

Edit: I just noticed that you said the same yourself. Great minds...

A really good developer can and will absolutely outpace a couple small teams of developers. The shortcomings is the domain knowledge is locked in one head. You gain and lose a lot depending on how many people are involved and how much you have to communicate.

Not having a backup solution is probably the single biggest shortcoming of the lone gunman in practice. The second is that the code is often not laid out or structured so that a team can work on it.

As the MapReduce example shows, the domain knowledge is not necessarily locked in one head in cases like this. I haven't heard complaints from Googlers about how Jeff and Sanjay write unmaintainable code, either. My own non-Google experience is that extremely productive programmers tend to write code that's more readable, not less readable.