My background is enterprise training, but I think there might be a few oranges thrown into your apple barrel, Nicole. If you compare one hour of platform lecture ("please hold all of your questions until the end") to an eLearning course covering the same content, the compression is considerably less. For example, I've found that simply converting a one-hour WEBEX'd recording of a product manager giving his product launch presentation results in approximately a 45-50 minute eLearning course.
Now, it is true that, if you account for Q&A time and instructor-informed anecdote, you will see even more compression. However, one design consideration is that a well-constructed conversion must account for the Q&A, either by bringing the most common questions forward and rolling the answers into the content (from experience I use 80/20 here, in that 20% of all questions are recurring and account for 80% of the Q&A period), or by providing a near-time feedback loop (email response, social networking chat feature, etc.). The result is that you only recover about half of the Q&A time, either directly or indirectly (you still have to account for the cost of answering near-line/real time questions, even when posed electronically outside of the elearning recording).
This means that the largest recovery is instructor anecdote which, while 100%, is rarely more than 5-10 minutes out of every hour. If you total up your time budget for a typical hour of training (averaged over a day long course), you get 5 minutes for comfort, 5-10 minutes for Q&A, and another 5 minutes for instructor "padding" (anecdotes). This leaves a worst-case budget of 55 minutes total seat time in an ILT environment. The best recovery you will see, though, is approximately 10 minutes (half the maximum Q&A time and all of the instructor anecdotes), reducing the committed learner seat time to 45 minutes, or slightly less than a 20% compression.
One caveat is that, in my earlier example of the product launch presentation, the compression there seems to be intrinsic to the conversion; it has nothing to do with anecdote or Q&A period. I concluded long ago that there is a pacing advantage to eLearning over ILT attributed solely to its modality (a fancy way of saying I have no clue why, but the outcome is reliably predictable). In my case, I have learned to count on this factor when forecasting. Conservatively, I would put this at about a 15% savings and, when combined with the above calculation, the net compression can be as high as 35%. Absent any other factors, though, I would say that banking on anything more than that risks a distorted forecast and a potentially bloated curriculum -- leaving the developer with trying to stuff 10 pounds of feathers into the proverbial 5 pound bag late in the development cycle.