Debating a "transitional" Python 2.8

Having one binary support both 2.x and 3.x code sounds good, but given the deliberately backwards-incompatible features of 3.x (especially the new, saner string/unicode handling), is it really feasible?

I mean, of course you can have something like a wrapper that detects the code flavor and selects the proper interpreter internally, but I think the interesting scenario would be having support for mixed 2.x/3.x code bases where only part of the code is ported / written to 3.x.