go1.3.3 although go1.4 will have this problem as well

When using the DefaultTransport, in our case it was via the ReverseProxy, the
ResponseHeaderTimeout will create a timer that is never closed.  The code relies on
expiring.
If you have a proxy under load and a timeout that is long in our case 15 minutes but
even anything in the range of 2-5 minutes, the number of timers starts to overwhelm the
Heap and GC.

In our simple runs under load, we can generate 10k req/s.  We see the heap grow from 2Mb
to 90Mb all due to timers.  The GC cycle starts at 2ms and grows to 120ms.

Our current workaround is to avoid setting the ResponseHeaderTimeout and rely on
DeadlineTimeouts.  Closing the timers created persistConn.roundTrip() would also
alleviated the explosion.