This thread has become pretty stale, but nevertheless I thought that sharing an additional experience with extreme splitting would fit better here than in a new thread.
For the first time since embarking on my extreme splitting adventure, I have used extreme splitting for census sources. I settled on the level of splitting for census data as being one Master Source in RM for each census page. I have been quite satisfied with the results. These are U.S. Population Schedules and I get my images from ancestry.com. Each image is attached at the Master Source level, and so far each Master Source has only have one image (a point I will return to shortly).
I place my census transcriptions into the Master Text area of the Master Source. I have chosen to use the Source Text area rather than the Source Comments area, but that would be an individual choice for any users who might wish to use similar techniques. I do not transcribe an entire census page if, for example, I'm only using the page as a source for one or two families. I only transcribe the families I'm actually using. For many years, I have maintained Web pages with census transcriptions, totally separate from RM. For each family, there is a transcription and a narrative. I'm still maintaining these separate Web pages. When I research a family in the census, in addition to creating the transcription and narrative on my separate Web pages I now copy and paste the transcription and narrative into the aforementioned Source Text area of the Master Source. The transcriptions look better on my Web pages than in the Source Text area because they are rendered in a non-proportional font so they line up correctly (I use the <pre> tag in the HTML). There is no way to accomplish the same thing in the Source Text note, nor in any other kind of note in RM. If there are multiple families I'm researching on the same census page (and there usually are), then I copy and paste all of them from my Web page into the same Source Text note in the same Master Source. It's really easy and fast to do so.
I wondered about the need or value to copy and paste my census transcriptions and narratives from my Web pages into RM. But I finally decided that without doing so, the linkage between the two was just too tenuous for long term data integrity (over decades or even longer). Plus, the copy/paste process is extremely easy and fast, and with extreme splitting it it only has to be done once. My census transcriptions also get copied and pasted into census fact notes. But they way they are formatted there is for display in narrative reports and is very different in format than what I am copying and pasting into the Source Text of RM's Master Source.
As to each Master Source being associated with only one census image, there are cases (especially pre-1850) where one "census page" is really two pages in the census book - the left facing page and the right facing page are really one big double page. On microfilm or at ancestry.com, one of these "double pages" appears as two images. Most people only look at the left facing page, but I find useful information also on the right facing page. So if you want to link the entire double page into a single RM Master Source, you either have to link in two images or else you have to convert the two images into a single multiple page graphics file. I do the latter, and convert the two images into a multiple page TIF file.
Finally, I have not processed any other kinds of censuses in this extremely splitting way yet, but I can't think of any reason they wouldn't work. They would include state censuses, censuses from other countries, and non-population U.S. schedules such as the agricultural schedules.