Interaction with the OS

Network
poll() / select() on a busy cache spend more time actually checking each FD than squid doing the actual network work
Squid does a lot of pre-buffering itself of server and client sockets rather than letting the OS handle it
Squid sockets are generally short-lived, and the default UNIX kernels have very high MSL lifetimes
In a lot of cases the data can be piped straight from server to client without going through the convoluted trip
Deferred reads waste a lot (15%) of possible read events, and should be removed and replaced with correct rather than lazy read scheduling
An event-driven network IO system would be perfect