Thank you for your question and my apologies for the late reply. One of our engineers looked into this and provided the following comments: if you only need the image, and nothing else from the page, then it probably would be a little more efficient to use hgRenderTracks. However, please keep in mind we do have limits on the number of hits to our server, see
http://genome.ucsc.edu/FAQ/FAQdownloads.html#download2. If you are running some CGI web-service thing where you cannot control how many hits and requests are coming in per second, it would be better for you to run your own local instance of the browser (our Genome Browser in a Box) and fetch from that. More info on the Genome Browser in a Box can be found here:
https://genome.ucsc.edu/goldenpath/help/gbib.html.