The answer is "it depends". Even if you're using all extracts on the data engine, your concurrent performance will depend on the size of the extracts, the mix of unused/used views, and a few other things. Additionally, the definition of "reasonable" performance varies widely from person to person: for me I draw the line at a few extra seconds, while some people are perfectly happy to wait 30s or longer for one particularly nasty view. If a viz that normally takes 3 seconds is taking 6, is that reasonable? So it's hard to give an exact number of users. There's also the variability of the ratio of concurrent to active users, which we usually estimate at about 10% but could vary quite widely for short bursts.
Sorry I can't be more help. My gut wants to say "oh yeah, you should be fine at 100 concurrent, even up to 200-300 on that kind of a machine" but I just can't assign much weight to that assessment given the complexity of this topic.
How much RAM will Tableau Server actually use? The server we are currently using has 24 GB in it, because somewhere along the line that was recommended to us. If it's a 32-bit app still, can it even use anything over 3.5-4 GBs?
The Tableau Data Engine is 64-bit software and can utilize lots of RAM. Don't expect to see very high RAM utilization though, other than under heavy load with large data sets: It doesn't aggressively pre-pull all data into memory, it does so only as needed.