Just done that testing on my localhost
Created a small course with a single topic, included an image with no language wrapper, some lorem ipsum dummy text and an image, with the dummy text and second image wrapped in a multilang div (using my multilang hack that allows divs as well as spans - see teh github link previously in this thread) and I installed a random selection of 11 additional languages.
Just English content
Average screen load time for this test content 0.06s
Content in 2 languages (filter shows one)
Average when changing language to a new language approx 0.11s, Average when changing to a language previously viewed 0.07s
5 languages (filter shows one)
Average when changing language to a new language approx 0.11s, Average when changing to a language previously viewed 0.07s
12 languages (filter shows one)
Average when changing language to a new language approx 0.11s, Average when changing to a language previously viewed 0.07s
So to me and in that limited quick testing on a localhost system there would appear to be no increase in system stress from the number of language used - there may be a minor impact of using the filter at all, but it does not appear to be determined by the number of languages being used.
My conclusion would be that the multilang filter option is certainly scalable to the extent you want it without undue impact on system resources.
The decision, then, should be based not on system performance but on development and maintenance time for the content itself and what suits your particular needs - for us that was definitely not setting up and maintaining separate courses for each language, for others that may be the best option. As a personal view, I can't see the benefits of maintaining separate courses at all - but I'm sure there may be a set of circumstances where it is preferable.
Hope that's helpful
Richard
PS. I've attached (as a txt file) the html I put in to a topic section to do this testing - you may notice that as I was rushing I was a bit careless with my copy and paste and the images being used are actually external links to another site rather than installed on my localhost - that may have some impact on page speed.
Also - my test involved having the filter on for both headings and content, which also may have an impact on speed over no filter or filtering just on content. I didn't run a page speed test without the filter enabled at all for comparison, but you may want to do that as well in your own tests.
R