What we're doing to solve it: ship small, free games that take a week or two to build and hope they get word of mouth. If they fail, no big deal, we learned something from making them and they contribute to our overall portfolio. If they succeed we don't make any money (since they're free) but they do generate a lot of goodwill and people notice and remember us for in the future.
We're trying something different with the next game by releasing a trailer early: http://contranoid.com. Let's see what happens. Adopting a more experimental approach feels more comfortable, either way.