Some months ago we ran a small experiment on the Beta site of a mobile application in which we ran a small 100-BU (Browser User) test to see how performance of the application changed as the load increased, linearly, from 1 BU to 100 BUs.
As a followup, we repeated the experiment on the same application -- which now has become a full-scale "non-Beta" application -- with the results shown here:
Further Comparative Improvement of Performance Under Load
As you can see from the charts, things didn't necessarily get better, and you could say that there was some deterioration. But remember, that later data -- from over four months ago when the tests were first run -- is now from a live application. Right now it appears this application is "good" for up to about 25 simultaneous users, but performance drops off quickly after that -- as much as a 5x degradation beyond 50 simultaneous users.