Profiling my scripts, the biggest drain on performance is the autoloader, specifically the require/include line.

A current site includes 38 files.

I did a quick test using include/require/require once/include once. Profiling just the autoload function and changing only the include/require line here's what I got just as a very basic test.

require_once 124.75ms
require 116.64ms
include_once 120.56ms
include 119.24ms

All very close. Nothing really in it. However, 38 calls to include/require are using 70% of my script's CPU time so it's a hefty chunk of resources and easily the most resource intensive function I have.

Now that still seemed a lot to me but perhaps it just a file I/O bottleneck. As a basic test, I tried:

PHP Code:
eval('?>' file_get_contents($file)); 
instead of
PHP Code:
require $file

The result... 24.29ms roughly five times faster. This is rather astonishing. obviously eval is bad™ but this proves that include/require are doing a hell of a lot more than simply loading the file. By using eval(file_get_contents()) I've improved my entire script performance by almost 100%.

Any idea why include should be so comparatively slow?

edit: the files being included are doing no real processing on include, they are all just class definitions.