While no agency that receives its annual budget from the president and congress can remain politics-free, the NSF's decision makers are career scientists and grants are decided by peer review without having to be cleared by government bureaucrats to meet some outside goal. Many other countries have adopted the NSF model for their own scientific agencies and I feel sorry for the ones that haven't.
In 1986, computer science research came under a new Directorate for Computer and Information Science and Engineer (CISE). NSF CISE provides most of the government research funding in theoretical computer science including my own.
The entire theory branch of the NSF, save program director Tracy Kimbrel, changes this summer and who those new people will be remains a mystery. MIT Engineering dean Subra Suresh is rumored to being vetted as the new director of the NSF. Susan Graham is collecting nominations to replace CISE AD Jeannette Wing. The search goes on for people to take over for CCF Division Director Sampath Kannan and Theory program director Richard Beigel. If you have suggestions for people, let me know and I'll pass them along. The future of theory funding depends heavily on us getting good people in those positions.
On the CCC blog, John King a notes how computer scientists are quite harsh in their grant reviews and how that can ultimately hurt the amount of CS funding (Ed Lazowska also chimes in). Probably comes out of our unique conference culture where program committees usually have a fixed number of slots to fill so tearing down other papers opens more room for papers you like. But NSF grants are not a zero-sum game. You shouldn't lie about the quality of a grant proposal, but neither should you look for reasons to tear it down.