LiteSpeed should not restart itself due to queued up slow connections. Post some log entries of the restarts. Check error.log and stderr.log.
If connections are queued in a big way, more cpu resources will be used however I still do not cannot see case where a "slowed" down server due to small pipe causes huge spikes in cpu and load.
A slow repsonse from the server, for whatever reason, can create a cascading effect where the end user starts refresh their clients thinking you website went down or something is wrong, causing DoS like effect where 500 accessing clients appears like 2500 clients. I imagine this to the case on your end.
Do you have logs of the litespeed restarting?
It could be a case you have some logic flaw in your php backend. If all php backend process, fastcgi/lsapi, are unresponses, they are restarted but a slow output to a client should not delay php as php sends all output to the lsws buffered. Once the php sends all the stuff to lsws, the php process's job is done. It is up to lsws to slowly feed the data to the client over the small pipe.