Do not develop one

Rather do not start with the development of a framework. The success of RoR is due to the fact that it is extracted out of a working program. If you look at frameworks like MFC/OWL/XLib you can see that they are developed before the application of those frameworks and it shows.

Use common idioms

One of the best frameworks is Unix system call library. A thread runs through most of the calls and it is very easy to memorize them. Look at open-read-write-close idiom - and the number of times it is used through out the library. If at all you need to develop a framework look around for similar framework/libraries and use the common idioms from them.

Use at least in two projects

If you are not using a framework in more than one project, it is not worthwhile to develop it in the first place. Application development is not easy. You do not need to complicate it by additional burden of making it generic also.

Create a framework out of working code

Agile practitioners say not to invest upfront in framework development. You can take it a little further. Do not invest in framework at all till your first application is ready. Then put in the efforts and your best resources to extract the framework.

Make the framework for a niche

Enough frameworks exist that try to please everyone and please none. You do not add one more to that list.

Recently I was reading a blog entry about ruby vs. lisp. One of the active lispers is so railed (pun intended) by the success of RoR that he even started a project “lisp on ladder” or some such silly name. (I am happy with this discussion as it made me look at lisp and enjoy it). IMO the basic point missed out is that Rails is not developed to create a 15 minute demo of creating a blog application. That is the effect of extracting a framework from a working project.