Unexpected EOF when pulling a large Docker image

After upgrading from 5.3.1 to 5.10.1 we are encountering an issue when pulling a large image from our docker repository. The image is 4.8GB in size and one of its layers is 3.15GB.

Downloading the largest layer lasts for about a minute, after which it is interrupted and the Docker client says it is retrying to download it. It tries to download it three times in total and afterwards gives up. This is the output:

latest: Pulling from android/agents/prod
16da43b30d89: Pull complete 
da8f233a88a5: Pull complete 
8e7d63e71a8f: Pull complete 
b29dcc02a811: Pull complete 
7ed6add7b5ec: Pull complete 
ce70d751a47d: Pull complete 
87e1d3619cc3: Pull complete 
2358ceb23f2c: Pull complete 
8d7bb7241e9c: Pull complete 
707f4b45f3aa: Pull complete 
134248cc1d42: Pull complete 
a7842e8cdd16: Pull complete 
bc492a4c956b: Pull complete 
8c0a95ad071f: Pull complete 
8d843f37765b: Pull complete 
4a4017d48cc3: Pull complete 
fc4d43079ec4: Downloading [==================================================>] 3.15GB/3.15GB
5680e5185b63: Download complete 
7468f2bf4fa0: Download complete 
016e8b573102: Download complete 
df79b4c97258: Download complete 
df609c7c4f32: Download complete 
27f658842903: Download complete 
864bb8fcccf1: Download complete 
unexpected EOF

Artifactory itself is running in Docker behind Nginx reverse proxy. Nginx error log shows this:

2018/04/12 10:45:38 [error] 7#7: *4002694 upstream prematurely closed connection while reading upstream, client: 10.7.201.207, server: docker.xxxxx.com, request: "GET /v2/ 

That leads me to believe that it is Artifactory resetting the connection, but I found nothing relevant in its logs. All timeout settings are unlimited as well as the maximum download size.

The image is successfully downloaded when connected to a fast network, but fails when connected over wi-fi and VPN, for example.

 We were able to download large images prior to the version update without problems.