Thanks for filling, could you reproduce this while logging a net-internals dump? Instructions here:
https://sites.google.com/a/chromium.org/dev/for-testers/providing-network-details

Also, something I noticed while testing it: it seems like the problem is degenerative.  If I start a new session and do a basic google search, everything works fine, but if I visit any site that uses a lot of data (say, watch a video) then it becomes unusable.  And not just that site but all web pages afterwards are affected.  

Also, I forgot to mention that once I try to use chrome over the SSH tunnel, everything else using the tunnel stops working.

I have emailed my log file to csharrison@chromium.org.

Hm thanks for the log.
I'm seeing the following SOCKS5 connect sequence in a bunch of places:

t= 4106 [st=    3]     +SOCKS5_CONNECT  [dt=30001]
t= 4106 [st=    3]       +SOCKS5_GREET_WRITE  [dt=0]
t= 4106 [st=    3]          SOCKET_BYTES_SENT
                            --> byte_count = 3
t= 4106 [st=    3]       -SOCKS5_GREET_WRITE
t= 4106 [st=    3]       +SOCKS5_GREET_READ  [dt=0]
t= 4106 [st=    3]          SOCKET_BYTES_RECEIVED
                            --> byte_count = 2
t= 4106 [st=    3]       -SOCKS5_GREET_READ
t= 4106 [st=    3]       +SOCKS5_HANDSHAKE_WRITE  [dt=0]
t= 4106 [st=    3]          SOCKET_BYTES_SENT
                            --> byte_count = 33
t= 4106 [st=    3]       -SOCKS5_HANDSHAKE_WRITE
t= 4106 [st=    3]        SOCKS5_HANDSHAKE_READ  [dt=30001]

Where the read is just stalled until the timeout happens.
mmenke@, any ideas here?

I bet the degenerative effect you're seeing is due to us accumulating a bunch of sockets that just stall. 

Hrm...  For that connection, we establish a connection, do the first round of the SOCKS5 handshake, send our half of the second round all in under a millisecond.  Then the server never responds.  We run into a connection timeout when we fail to negotiate a connection with the proxy in 30 seconds.  Really doesn't look like an issue on our side of things there.  We're just not hearing back from the proxy.

Could imagine learning that since the first round time was under a millisecond, the second one should be, too...But that would just be working around whatever underlying issue is causing us not to hear back from the proxy in the first place.

Not sure why you're only seeing a problem with Chrome.

Issue has not been modified or commented on in the last 365 days, please re-open or file a new bug if this is still an issue.

For more details visit https://www.chromium.org/issue-tracking/autotriage - Your friendly Sheriffbot