The notion that undecideds always break for the challenger is an UL.
The data is very muddied, but what pattern there is shows undecideds are slightly more likely to break for the incumbent.
The pattern has been clear since the 1930s. The only exception was Reagan vs Carter, in which it broke for Reagan.
Undecideds break for the incumbent, or the incumbent party.
it depends on the unfavorability #. I expect the undecideds to break right along the favorable/unfavorable # for Bush - what is is running at right now?