I don't buy the comment about firewalls lowering security by making them more complex. When you consider the default build of Windows or a comon *Nix, or even the typical corporate build, you'll see the system is already insecurely complex ipso facto.
So first you say that you don't agree with the idea of increased complexity reducing security... and then in your very next statement you say the systems are already "insecurely complex". You sound confused.

What do you have to lose by adding something that could potentially close avenues of approach or attack?
How does an internal firewall close avenues of attack? If you have a filter segregating that network segment what points of attack are you worried about?
Attacks from the outside will be dealt with by the external filter.
Attacks from host based malware can be prevented by disallowing in installation/execution of unsigned executables.
Internal worms will use the same channels as internal trusted communications so a filter again will not work unless it is integrated with malware detection which needs to be maintained.
So where are these attack avenues?

As catch and his oompah loompahs are always reminding us, secure systems, designed properly from the ground up, don't need the level of attention for patch management, firewalls, and security applications.
From the ground up? Following basic security principles... (control what users do, control what code can be run, control what services are used) and suddenly these internal firewalls become a non-issue.
Would you run a personal firewall or AV system on Linux? Why not? Because people tend to use normal, non-administrative accounts. There is no fundamental difference in structure that allows these attacks against Windows and not Linux.

I've said it before...Information Security is about defense in depth.
I've said it before... "defense in depth" does not mean doing the same **** multiple times in slightly different ways. Think of mantraps... you want as few very high assurance bottlenecks as possible and nothing else. Each checking different things. Firewalls, network guards, reference monitors are all fine examples.

"Why have host firewalls? I have a perimeter firewall already!" What if the perimeter firewall fails?
Then it fails...
Given resources of X... what is less likely to fail... one very high assurance firewall that uses all of X or 3000 bad firewalls each costing 1/3000 of X.
Unless each of those firewalls is configured different what is gained? Why would the failure of one not pass to the failure of others?
If your primary firewall is compromised do the host firewalls even matter? Or can an attacker easily subvert the hosts through traffic control?

You can make your failure complete by having zero internal stop gaps or defenses, or you can have a fighting chance with internal measures layered upon other measures.
No one is suggesting this... different controls need to be used... the first concern is detection of the failure... then host hardening, etc. More firewalls is not defense in depth.

But please, feel free to put all your egg's in one basket.
Security is about assurance... not a pile of crap.
Single points of the highest affordable assurance are always better... the same reason a reference monitor is the ideal way to handle OS security.

Why you got positives is beyond me... the community is just full of people who have no concept of basic math skills. Damned American public schools.

cheers,

catch