1) Minimally, but as always, it depends on the application, the data model, and how the shared object in question was built. Best case is that it won't affect performance in any measurable way. Worst case example - someone creates a straight table with many columns and expressions, and hundreds of people use it simultaneously via the AJAX interface. That would cause serious performance problems. But that's not a "shared object" issue per se.
2) I have not heard of this as an issue. You may have stability issues with a cluster on SAN with lots of latency, but that's not specific to shared objects.
3) Is there a specific bug number you're referring to? You should contact QlikTech Support and either get an updated status on a specific bug or file a new one if it hasn't been reported before.
4) If you allow a user to share objects, there's no way to limit multiple copies of the same object that I know of. What is your concern?
Thanks Gary for your prompt response!
2) I saw one discussion about .SHARED file corruption and QVS stability, which I cannot now find . Sorry.
3) From the 10SR4 release notes, I think it would be either 40652 or 37757 http://d1cf4w4kkla6tb.cloudfront.net/qlikview/10.00/9282/QlikView%2010.00%20build%209282%20SR4%20Update%20Release%20Notes.pdf
4) Concern is that QV allows you to drag over ("insert" shared object) the same chart more than once. I have seen a case where the user had four copies of the shared object chart on the screen at once -- QVS was churning hard to calculate all four simultaneously.
2) I have heard about shared file corruption but I haven't experienced it myself. My guess is that with a large number of users, and in a clustered environment, your odds of encountering it are increased. You can always make frequent backups of .shared files, which is a best practice anyway. You can also check out the QlikView Power Tools which has a shared file viewer if your shared file becomes corrupted.
4) I'm surprised you were seeing that significant of an impact of duplicate shared objects.
Now you have me wondering - is there any difference in the way QVS.exe calculates shared user objects versus baked-in server objects? I don't think there is, but I'm going to ask some people. Sorry I don't have any concrete answers.
4) The impact comes when you have an application with a wide table with 380MM lines and the chart in question is a pivot with 4 vertical dimensions and 2 horizontal dimensions (each dimension having a potential 2 to 81,000 unique values) and roughly 50 expressions. (Yes, this is a ridiculous chart.)
I'm curious to see what you find out. Thank you for sharing your time.