The lines for peer *.3 and *.4 contain an extra '0' in the options area
after ICP-port which may be preventing Squid from loading this config.
Other than that nothing is visible from the details you mention.
Some random possibilities:
* Check that all peers are detected as ALIVE by Squid. Peers that are
marked as DEAD (10 consecutive failures occured) are skipped in the load
balancing algorithms. That 'no-query' disables the UDP checks used to
detect peers going from DEAD to LIVE status, so error recovery will be
* Check that the "requests" you are referring to are not all inside a
single CONNECT tunnel. Tunnels count as a single request in plain-text
HTTP no matter how much 'crypted traffic they contain.
I assume this is the same 3.5.22 you mentioned using in threads from a
few days ago.