Squid performance lagging Apache

[Matthew Von Maszewski <matthew@xxxxxxxxxxx>]

I have a test bed using siege.  The test runs 23400 hits against 19  
static http items.  Squid, delivering all from memory, runs roughly  
75% of the throughput of raw Apache or Apache proxy.  The siege  
"client" box has plenty of ports so that TIME_WAIT is not an issue.   
The client and server have a direct 1Gbit ethernet wire between them.

The boxes are multi-core Xeons at 2.6G with 8G ram, running Debian  
Etch Amd64 (2.6 kernel).  The build has epoll() enabled.  Single core  
hits roughly 80% with squid.  With squid tests, a third box was  
running Apache to supply http content but not accessed due to caching.

Any suggestions as to what might raise the throughput of squid to  
approach / exceed apache2?  Any recommended profiling techniques?

Note:  I later configured 2 squid instances on the same box and  
divided the 23400 evenly between them.  The throughput was constant,  
same as if one instance.  CPU cores hovered around 37-40% during test.

Summary:  plenty of ram, disk not used, extra CPU power ... seems  
like an issue of socket throughput ...

Thanks,
Matthew