I've been reading asyncore lately, and feel that it's showing its age. 
Most loops of this sort (we developed something similar for ILU, about
15 years ago) contain handlers for timers and work tasks, in addition to
input handling.  For timers, typically there's a list of tasks and
times, often with a repeat period.  A system timer is set to the time of
the next task to fire, and the select() loop is exited when it fires. 
The loop handler then looks down the list of timer tasks, and executes
those ready to run.  Similarly, most loops of this sort include a list
of work tasks, and a policy for executing them, such as "take one task
from the front of the list and run it, then do the select".  This allows
background tasks to get run that don't have associated input or output fds.