Although there have been many great suggestions on moving to "open source" software, I'm going to make a slightly different pitch: that the government should create a pilot program to reduce the amount spent on costly licenses and reinvest that money in freely licensed software products that answer government needs but also may be applicable to the private sector. This isn't an appeal for a movement to "open source" but rather an appeal to fundamentally change the way the government approaches the acqusition and delivery of software as a business function.
The first step of the plan is to require all government agencies to reduce their costly licensing by at least 10%. For example, if your agency awarded $42 million on licensing last year, the agency would be required to set aside $4.2 million for this pilot effort.
Once the funding is identified, the next step would be the application of these funds to companies and projects focused on software that is freely licensed. This does not necessarily mean open source, but it does mean that the software could be used at any time, by any one, at no charge. These projects would apply for funding to meet tasks that support the operation of government (e.g. messaging and communication, auditing, data storage, et cetera). As the project progressed, not only would the results be freely available to the entire U.S. Government, but also to the public at large. In this way, the government is answering its own needs as well as assisting in the reduction of software licensing expeditures in communities and local governments across the United States.
The final component is one of metrics and information dissemination. A website available internally and externally would promote the pilot program, allow for electronic submission of applications, follow progress of accepted projects, and provide trusted download points for distributable versions of the software (likely multiple mirrors for the public, and internal mirrors for trusted government use). At the end of the pilot, the team in charge would be able to produce substantive data on how the funding was spent, what was created out of the funding alloted, what the level of adoption was inside and outside of government, and highlight some success stories. If the pilot is successful (a measure that should be defined), then the next year the effort would be increased to 20% of all costly licensing awards, and so on until such point that a majority of the commodity software used by the government for its tasks is freely licensed for itself and the public that it serves. The end-game would promote innovation, small business stimulation, and be deliverable practically anywhere in the United States.