App Tunnel throughput problems

I've noticed some strange issues related to throughput when using a java app tunnel. In my setup I'm using Windows 10 guest within VMWare ESX. For test purposes I created a 25mb file with random data, that can be downloaded via the app tunnel. Both the BIG-IP (v13 HF2) and the Windows 10 guest are on the same LAN.

When initially downloading the file by using curl I got very poor download speeds (around 200k). See example below.

>curl.exe -0 -o test.bin http://localhost/F5/rnd_file25mb.bin
  % Total    % Received % Xferd  Average Speed   Time    Time     Time  Current
                                 Dload  Upload   Total   Spent    Left  Speed
100 25.0M  100 25.0M    0     0   172k      0  0:02:28  0:02:28 --:--:--  213k

However, I managed to get 'normal' download speeds via de app tunnel, by first downloading for example an ISO. While the ISO download is running, but also afterwards. The throughput of the app tunnel is normal. I'm now getting download speeds around 3500k. See example below.

>curl.exe -o test.bin http://localhost/F5/rnd_file25mb.bin
  % Total    % Received % Xferd  Average Speed   Time    Time     Time  Current
                                 Dload  Upload   Total   Spent    Left  Speed
100 25.0M  100 25.0M    0     0  3501k      0  0:00:07  0:00:07 --:--:-- 3675k

I wonder if the poor app tunnel throughput has something to do with shortage of entropy on the Windows 10 system. I can imagine that the applet that is used for the app tunnel might need to do some kind of randomization.

On the other hand, it could be something totally different. Anyone experienced similar problems? Any tips on how to fix this?