FIPS certification is only available for systems that implement modest key lengths. Many of the approved algorithms are designed to support much greater key length, but longer keys are not allowed by the specs. FIPS won't certify 'em. It's a pretty safe guess that the allowed key lengths are such that the NSA can break them if needed using custom hardware or whatever else quasi-unlimited money can buy. Remember 20+ years ago when the gov't regulated all crypto as a munition? They still allowed low-bit encryption because they knew they could break it. They're still playing that game, except now it's done with standards and certifications instead of laws.
You really don't want to start making up your own ad hoc crypto. Approved algorithms have been extensively vetted by honest experts; any possible weaknesses would be very, very subtle. Using approved algorithms with non-standard "ridiculous" key lengths is probably the safest workaround to suspected weaknesses until... on second thought, key lengths much greater than the gov't "recommends" will always be a good idea! Keep in mind that any weaknesses in crypto algorithms would merely make them easier to break, but breaking still requires huge resources and takes time. Longer keys kick up that effort exponentially to the point that very probably nobody can break them in a useful time frame, provided that implementations are reliable and trustworthy.