It's hard (in an engineering sense) at that scale, but certainly not impossible, and easier than a lot of engineering problems the world has solved. It's harder because DoD is actively being attacked, but easier because they have a near-infinite budget.
The thing which makes it hard is humans, politics, and economics -- there is a huge amount of CYA with respect to vendor choice (hence, they're a huge Microsoft/Cisco shop), lots of little fiefdoms, an "up or out" promotion policy combined with people being in leadership roles for short periods (with minimal prior background), and lack of real accountability.
The Microsoft-ness isn't enough to kill them on its own; look at the Israeli military, which is also heavily Microsoft based, and has world-class computer security.
Use-after-free vulnerability in Microsoft Internet Explorer 6, 6 SP1, 7, and 8 on Windows 2000 SP4; Windows XP SP2 and SP3; Windows Server 2003 SP2; Windows Vista Gold, SP1, and SP2; Windows Server 2008 Gold, SP2, and R2; and Windows 7
"Microsoft thanks the following companies for working with us and for providing details of limited, targeted attacks against customers of Internet Explorer 6:
Google Inc. and MANDIANT;
Adobe;
McAfee;
French government CSIRT (CERTA)"
I think for single-purpose machines (like "control a UAV"), custom hardware makes a lot of sense, even for commercial operations. Unfortunately custom hardware usually ends up being a Windows box in a weird case, with some buttons connected over...USB.
A requirement that all components of the TCB be FIPS 140-2 level 3+ for anything which is routinely used in combat operations would please me, I think. Right now that's just for the crypto modules themselves.
eye roll