Comment by toomim

Yes, the limit is implemented in browsers, but it's for the protection of servers. The idea is to prevent a web app from DDOSSing a server somewhere.

This is especially a problem for servers implemented with one thread or process per connection, rather than an async event queue like nodejs, because each thread takes up a few megabytes of memory, and if a browser opens 1000 connections, that's 2 gigs of memory right there.

But if a server is implemented properly to support long-lived connections (by using node, or greenlets, or other async system) then it should be able to opt out of that browser protection. A standardized header would allow a server to opt out of it.