It seems obvious to me that the task isn't limited to an historical
research of timezone shifts, which is, all in all, a tremendous job
itself, but to associate a 'place' to each timezone shift; a 'place'
being either a city or an administrative subdivision like a county, a
region or whatever name is used in each involved country.
IMHO, in order to set off such a task, a free and common GIS should be
available to all the contributors, e.g.
http://postgis.refractions.net/,
initialized with the suitable administrative data (
http://gadm.org).
A TZ file is already associated with a country or a TZ domain, but
this TZ domain may be subdivised into quite a few local exceptions,
which could be recorded in the same db as the GIS.