We have a similar user base. Let's see... 440 distinct users ever, 238 distinct users in 2010, and 150 distinct users so far in June. I'm doubting that we actually hit 50+ concurrent users, though. And I'm afraid I wasn't involved in setting up the server and how it's handling load balancing.
There are QlikView customers VASTLY larger than us, though. I doubt that the QlikView Server software is going to be your bottleneck.
I believe that QlikTech recommends that you set aside an additional 5-10% RAM for every concurrent user. So if 8 GB is enough to handle one user, then let's say you need (100% + 7.5% * 50)*8 = 38 GB of RAM. Though you'd have to have documents much larger than ours to need that much. I believe QlikTech also recommends 2-4GB per core, so you're probably looking at a 10+ core server to handle the load. Again, your documents may not be as large as what I'm estimating with here.