NickLightbody has done some in-depth research in testing environments.
What I have found in practice is that FMS13 was ok. FMS14 was a big improvement and many times faster than a FMP client. FMS15 is a slight improvement over 14, but not as big of an improvement as 14 was to 13.
The biggest improvement is that the whole FMS is performing better with less memory usage. Therefore you don't need a super machine to run for several WebDirect connection at the same time.
There is as bigtom wrote a small improvement in speed also. You
dsBenchmark free from http://deskspace.com/downloads.html enables anyone to undertake this type of testing. You need to open a new WD user session in a number of different browser tabs, I use three browser sessions each with 5 tabs which simulates 15 WD users.
dsBenchmark will then display performance in a dynamic chart in each WD session.
To assess resource use you need for each load level to record the RAM use for each of the FMS15 processes, using the Activity Monitor on the server, including the additional Java processes that WD initiates.
This enables you to build a spreadsheet of the resource use which in the case of 14 was about 32mb RAM per WD user.
Do remember that it is essential to never set the FMS cache to a bigger number than required, start with 512mb and only increase if the Admin Server Console stats on cache hit fall below 95%. Allocating FMS cache prevents that RAM from being used by FMS processes (other than caching data), Java, the OS, and everything else.
Please then share what you observe with our community!