Expression Engine updates causing HTTP 503 errors

We have a problem with our Expression Engine sites (at least 3). This has been the case since I have worked on the projects for over a year, and is apparent in versions EE 2.7.3 through EE 2.8.1.

Here is how to replicate the problem:

1. Have a user (USER#1) log in and start updating content (either creating a new template, posting a new channel entry, pretty much anything that involves updating the EE backend MySQL database table).

2. While USER#1 is making changes, have a second user (USER#2) try to access the front-end EE-powered content.

Our sites get (for USER#2) intermittent 503 HTTP responses (site temporarily unavailable). This manifests as one of:

1. Full blown 503 HTTP responses (site temporarily unavailable).
2. CSS not rendering (so the site layout looks messed up). This is because the CSS files are returning 503s
3. Images not rendering because they are returning 503 errors.

Once USER#1 finishes making changes in the EE admin interface and the database is once more synced, USER#2 no longer gets any front end errors. This is a significant problem for us where we have 20+ members who post content to the EE admin interface across the United States (east coast through west coast)

Note that turning on debugging in EE won’t help because seems to be a low-level Apache/MySQL/PHP problem (hence the Apache errors - not EE or PHP), most likely in Code Igniter (underneath EE). Info about the server:

Server: Red Hat Enterprise Linux 6.5 Apache: 2.2.15 MySQL: 5.1.73 PHP: 5.3.3 EE: 2.8.1

My Apache env:

Max Requests: Per Child: 10000 - Keep Alive: on - Max Per Connection: 100 Timeouts: Connection: 300 - Keep-Alive: 5

Has anyone else noticed this behavior? Any ideas how to troubleshoot? The only way I can think of to troubleshoot would be to write a script that sends multiple cURL GET requests (with session ids) to the EE admin site (basically duping EE into thinking that a user is making changes to the content), and then have a traceroute (or dtrace) running on all Apache child processes looking for a bad response. However, that is a pretty sizable project which I don’t have time for. I am hoping that someone with more EE/CI experience will pipe up and say something like “hey - didn’t you set the foo bit to bar in the config (either EE, CI, PHP or Apache)? that solved it for me”

I know this is a magic bullet ask. Thanks in advance.