I think it raises a fair issue and I hope they're aware of this issue if it is as the link in the OP suggests. It seems to have only presented itself now that the server load is unreasonably high. If it has to re-establish a connection with the server every so often then it makes sense if it hits its cap then it disconnects from the server. I'd suspect this is in place to prevent any server connections from hanging, but I am not knowledgeable enough to suggest a better alternative to that problem or why it may work that way. Or how big the solution to this is or what problems might be caused by a fix. It'd also perhaps explain why when I get a 2002 and reconnect and it retains my position in the queue because it retains that position on refresh but the server may refuse the connection if load is too high.

The fact they've managed to increase the cap to 23,000 per data center by using their backup development servers I think should alleviate the problem, because it's less likely to hit that cap when it has to drop and reconnect.