On Mon, 17 Nov 2008, Daniel Stenberg wrote:

> My current suspicion is that the problem is related to libcurl's connection
> cache and the fact that both connections are done to the same server.

Here's a quick test for Andreas and Markus, as it seems to at least reduce the
chances for this problem to happen for me. This fix is not a final one but if
it does make a difference for (any of) you at least it is one step forwards:

diff -u -r1.771 url.c
--- lib/url.c 5 Nov 2008 21:46:40 -0000 1.771
+++ lib/url.c 17 Nov 2008 09:35:29 -0000
@@ -2565,7 +2565,7 @@
        }
      }

- if(match) {
+ if(match && !check->inuse) {
        if(!pipeLen && !check->inuse) {
          /* The check for a dead socket makes sense only if there are no
             handles in pipeline and the connection isn't already marked in

-- 
  / daniel.haxx.se