Page caching performance has regressed by up to 20% with no opcode cache

Follow up from #1064212: Page caching performance has regressed by 30-40%.

I haven't fully tracked this down yet, and this needs benchmarks to confirm once #1064212: Page caching performance has regressed by 30-40% and possibly #1064882: Remove syscalls from calling date_default_timezone_set(), date_default_timezone_get() are in.

Things to look at:

1. We include lock.inc in drupal_bootstrap_variables(), whereas we could get away with doing that only if we need to actually use the lock system. However I could only see a 0.5% (one or two requests per second) improvement from quick testing. Maybe there's other crap we're including as well though.

2. We can possibly claw back a couple of milliseconds (which makes a big difference on cached pages before anyone starts..) via #1055856: Optimize settings.php discovery. This can't be the cause of the regression, but it's worth looking at compensating for it from a different direction.

3. I'm wondering if we could change the bootstrap_modules cache entry to a variable in system_list(), this would be quite hacky but it'd save a query for both cached and uncached requests.

4. It's possible that bootstrap.inc just got really bloated with quantity of code, but given lock.inc doesn't make much difference either way I don't think that's too likely to be the problem.