This Victorian Electoral Commission newsletter from May 2009 is a few years old now, but I remembered an article on the back page, relating to geographically plotting the locations of addresses on the electoral roll (with Kororoit illustrated as an example). It's a bit like that racial dot map that was doing the rounds of here and Facebook a couple of weeks ago, but instead of the colour of the dot representing ethnicity, it represented a particular polling booth.
The point I am trying to make is that although there is no compulsion to vote at a particular polling place/station/precinct in the manner overseas, there are very definitely booth catchments, that become obvious when looking at this map. This implies that your rough intra-electoral boundaries representing booths, while possibly not perfectly aligning with booth catchments, will certianly be "somewhat accurate" and I think very useful. They will give us a good idea of areas of voting strength within electorates.
See also
this website, which shows voting results in 2010 by polling booth for specific electorates, and also wider afield: "Melbourne Metropolitan", "Sydney Basin", etc. He uses dots to represent the location of booths, with larger dots for larger electorates, and shaded in a few different tones (less than we use here, but still better than a simple bi-chromatic "either red or blue" scheme).
You may also recall this map, that is in the Gallery, and that I've posted before:
I have it for the whole of the Greater Metropolitan Area (881 polling booths in total). It is presently just the 2010 state election results and booths, but I'll probably upload it here sometime after the state redistribution is finalised. I may start work soon on doing something similar for the federal booths.
Additionally, the federal Parliamentary Library prepares vote maps of electorates by CCD/SA1 (small statistical areas used by the Bureau of Statistics during the Census). My understanding is that the AEC provides details of how many electors from each CCD/SA1 vote at each polling booth, and the Parliamentary Library then creates the maps through weighted averages of booth results. The only problem is, each map is then shaded into... I don't know... quintiles, I guess - five bands, anyway, but each electorate has its own scale, so one map is not directly comparable to the next map.
