Intermittent SSL negotiation errors

[Michael Nachbaur]

I'm developing an application in Apache2/mod_perl with nginx sitting  
in front of it in a reverse-proxy configuration with SSL/HTTPS on  
nginx.  We have multiple instances of this configuration installed,  
and they communicate with each other with Ajax requests called via  
LWP.  So an HTTP call to one web app can result in that webapp  
connecting to the nginx on another server in order to make an Ajax  
call to the other machine's nginx/Apache.

This works in almost all cases, though on occasion we get a 500 SSL  
negotiation failed message, and get these error logs on the "target"  
instance of nginx:

==> /var/log/nginx/admin-access.log <==
10.18.3.216 - - [18/Jul/2008:17:27:11 +0000] - "400" 0 "-" "-" "-"
10.18.3.216 - - [18/Jul/2008:17:27:11 +0000] - "400" 0 "-" "-" "-"

==> /var/log/nginx/admin-error.log <==
2008/07/18 17:27:11 [crit] 3094#0: *2302 SSL_do_handshake() failed  
(SSL: error:140760FC:SSL routines:SSL23_GET_CLIENT_HELLO:unknown  
protocol) while reading client request line, client: 10.18.3.216,  
server: _

==> /var/log/nginx/admin-access.log <==
10.18.3.216 - - [18/Jul/2008:17:27:11 +0000] - "400" 0 "-" "-" "-"

Has anyone seen this at all?  Any light you can shed on this would be  
greatly appreciated.