Apache has got MPM (Multi-Processing-Module) which has got many implementations, but mostly people are using 2 models, either prefork model or worker model. Prefork model is based on creating process as per request comes and the all the existing processes (as a pool) are like shared socket of master socket, So the delegation of request to any respective process is been handled by kernel. Worker model has got threads to handle the request. Prefork is stable in terms of No multi-threading so no clash or ambiguity of any shared library of Apache, though any unknown library can behave erroneous in case of worker model. But at the same time prefork model consumes more memory than worker model.
Using PHP, people suggest to use worker model, in case you are using AJP and then Tomcat, people suggest to use Prefork.
Configuring Prefork MPM for handling more number of requests
< IfModule mpm_prefork_module >
Max value of MaxClient can go to 256 untill you set ServerLimit also, but once you set ServetLimitm you can increase MaxClients, in general people have reported that Apache can handle max of 4000 requests after that it dies.
You should reduce KeepAliveRequests in apache, because this cause more memory usages by Apache and set the KeepAliveTimeOut for sure.
If you are using AJP protocol to connect tomcat, you should configure AJP connector, if you are using HTTP you should configure HTTP connector. In case of mod_proxy, Apache will redirect the request to tomcat in HTTP form, in case of mod_jk Apache will redirect the request to tomcat usin AJP.
< connector port="8009" maxthreads="512" minsparethreads="25" maxsparethreads="75" connectiontimeout="10000" enablelookups="false" redirectport="8443" protocol="AJP/1.3">
< connector port="8181" maxhttpheadersize="8192" maxthreads="50" minsparethreads="10" maxsparethreads="20" enablelookups="false" redirectport="8443" acceptcount="100" connectiontimeout="10000" disableuploadtimeout="true">
And you should make sure that you are able to serve the request from tomcat as fast as possible, I have seen as google give good value, if you can serve request in less than 80ms.