I don't know if it would require a big budget, but it would require an awful lot of research and testing. Since this would take a lot of time the target Debian release would have to be Wheezy, not Squeeze.

I see the following problems:
1. There are most likely some binary blobs in the kernel which aren't version independent. That alone could be a show stopper.
2. The change in the kernel version's naming convention.
3. The power draining problem introduced in 2.6.37. I'm not sure if it would even affect the N900 platform but if it does even ea working updated kernel would be useless.
4. All those closed applications in Maemo, half of them probably written in a poor static way that breaks if a version number in some remote corner of the system changes. And no chance to teach them to use the new version or to fool them so that they think they still have the old version.

It's always the same with abandoned closed systems. This is why I'd be willing to pay somebody who can solve these problems and put Maemo on more solid, truly FLOSS ground.