OK, for my next trick, I pulled a lion outta my hat.
Musta been the wrong hat.
(I don't know why a Bullwinkle reference suddenly popped into my head, but there we have it.)
Then I went back to testing this.
Just to eliminate as many variables as I could think of, I went for an extremely simple example of transclusion. I just created a series of integers from 1 to 19200, each separated by a single space, and created a page
here. I then jumped over to the O-portal and transcluded that page
here.
Character counts:
Text entered: 104093
Text rendered: 104101 - <p> and </p> tags and /n (or whatever the proper notation is in this case - an actual newline, not a visible tag) added by rendering
Text transcluded: 65550 - <p>< br /> and /n added at the beginning, <p> from first render retained, all text from the number 12774 on (including the first render's </p> tag) dropped
If we consider all code (including the <p>,</p>, and /n) from the first rendering as having been transcluded on the second render and subtract the characters added by the second rendering (<p>< br />/n</p>/n), then a total of 65535 characters were transcluded into the O-portal.
(If you're counting along at home, I have to use put an extra space after the < in the br tags because the forums will translate it into an actual line break if I remove that extra space.)
I think it's pretty safe at this point to say that we have an effective (if not exact) transclusion limit of 2^16 characters.