I have had similar issues and the end result on many are the http server isn't tuned to handle the load being placed on it.

How many total http process are there?

There is a parent and it spawns off a number of child processes. You can elect to have the children start at startup or get started dynamically based on need. I think the default is to have an intial number of children already running.

I never adjust anything until I understand the load being placed on the http server...the bigger picture rather than the detailed look you are doing.

------------------------------------------------------------
netstat -an ....

I look at the state of the send and receive buffers of all the clients (netstat -an | grep 80).

Look at how many are established and if any or in other states. Look at the Recv-q and Send-Q for each connections and see if they are stuck...or what those values are...0 0 means that particluar connection doesn't have any transactions in progress. If you see one of the buffers with a value that isn't changing, that particular connection may be hanging.

------------------------------------------------------------
access.log

I also go through my http logs and look to see how much data and the fequency of of the requests. So you can look in the access log and see the ip addresses of clients, their frequency, and the size of the data involved in a get for example. On one occassion we had someone using the server as a patch server (the clients would connect over http). I grep'd through the logs and counted the bytes over time and eventually found 12 clients hitting the server (http) with high frequency and large data requests and the http server just couldn't handle it.

grep through your access.log for ipaddress, gets and puts, and the amount of the transfer...see the format of the access log to determine the fields of an access log entry

So...after such an analysis I was able to determine the frequency of when the http process got to 100 percent and I exampled the
data accumulated during this frequency and then I stopped some of the clients I saw making frequent requests to test my theory.

---------------------------------------------------------------------------------------------------------
How to solve or provide relief for overworked HTTP server...

So once you understand if this is a load issue, that doesn't mean you automatically start tweaking the httpd server. When I saw the same clients frequently hittng the http server I reconfigured those clients to use something else. I could also have elected to do a hard-temp fix by restarting my http server every so often. These things are easier to do if you're too busy to get to the perfomance adjustments you need to make to resolve an overworked web server.

In short, I counted the amount of data that was sent through the webserver each time it went to 100 percent and I knew how long the webserver would run from startup of the webserver to the time of the 100 precent cpu usage.