Well. I'm bored. Noone's calling. Let's give my 5 cents
I'd say there are three major types of holes
1) Holes that are in standard services. I.e. in apache that comes bundled with osx, or in the tcp/ip stack.
2) Holes that are in an application/framework (Ie, you visit this evil-site on the net that has a malicious script of some kind) (NOTE! "stupid" users clicking "Yes" To install "This 1337 software" does not count in this category).
Interesting in this category is like, any program counts. Say, if somoene found a hole in the unix cmd "ed" (old-wack-editor, very very few people use it), that could say, make it execute arbitary code (bufferoverflow), it'd count as a "security vulnerability" in the statistics. However the number of affected people would be minimal. However, this code wouldn't allow you to gain "control" of a computer, unless it was the 'root' user that used the editor. Several UNIX systems being opensource, People find these kinds of bugs by simply scanning through the code (Belive me, people do this, for fun (...)). Also, I'm not sure how US gov.t. Count, do security holes found in pre-release software count? I.e. If a pretty big Opensource utility found a security hole only present in pre-release versions. I'm sure they'd present it anyway, as (some) security minded people might be using it. Also adding to the statistics.
hm. I kinda lost myself in my argument. Can't recall where i'm heading..
3) "Holes" that really are users not being prevented from doing stupid things (Ie, installing some odd program). Media seem to like to focus in this cathegory..
ohwell. bottom line. I should read the article from US.govt instead of bitching.
I'd like to know what kind of holes they count.. and how many of them comes from code written by a linux hacker studying in highschool

(No offence, everyone's been there ;D )
balh, just ignore me.