I'm a big believer in avoiding the Ad Hominem fallacy, so even though Dan O'Dowd has every reason in the world to make this up from whole cloth to protect his business, lets treat the claim seriously and address it.
It is true that evil foreign agents could try to get trojan horses into the Linux sources (as could evil domestic agents). However, they could also get jobs with companies like, say Green Hills, or other defense contractors. The latter would seem like a far more direct route to sabotage, since you get a close look at how your software will be used and thus can plan your sabotage much more effectively.
Although it is true that people working on defense contracts usually have security clearances, it is far from clear that such clearances actually prevent espionage or sabotage. I know of no studies that validate the methodology used in security clearances, and certainly the "security clearance" barrier hasn't prevented lots of folks from causing damage to U.S. interests even when they've had the clearances.
It is also the case that much of the software that goes into defense systems is produced by people with no clearances whatsoever -- I doubt that Green Hills, for example, always goes through the trouble of clearing the guys who work on their base software products if they are not going to be doing classified work.
We also have the question of the "many eyes" theory of open source security, which O'Dowd makes fun of. Many open source advocates note that since anyone who wants to can read the source code to an open source product, it is harder to conceal back doors. O'Dowd attacks this by saying that there are none the less security holes found quite regularly in Linux. What he does not mention is that there are also security holes found quite regularly in Windows and other proprietary operating systems, and that there might even be security holes in his own products. The question we are looking at here is not whether or not there are bugs -- the question is whether it is easier or harder to conceal an intentional flaw in an open source system.
Although it is true that the ability of large numbers of people to read the code is no panacea, it certainly is a help. There are comparatively few people who get to read the code in proprietary systems, such as the ones Green Hills sells, so there are fewer people in a position catch a trojan inserted by a rogue programmer.
Mr. O'Dowd also misses one of the most important aspects of security -- he fails to discuss the economic tradeoffs (if any) being made in a given security decision. He mentions only the possible problems of using an open source operating system, but he ignores the price associated with not using one. Against the weak claim of decreased security, we have to balance the loss of functionality and increased cost that using a proprietary operating system might cause. Developers do not select open source software at random. They adopt it because it gives them better functionality and has a lower cost.
Indeed, the cost savings and productivity benefits of open source systems might easily make it possible to devote more effort to security in a design, and the improved tools available can make security far easier to implement. Open source operating system users take features like packet filters, MMU based memory protection for multiple processes, logging facilities, etc., for granted, but these features not available in many conventional embedded operating systems. Even the ones that do have any particular feature rarely provide the breadth of functionality of the open source systems.
Lastly, let me note that Mr. O'Dowd appears to be inventing the threat he describes. I doubt he has any actual evidence of evil foreign agents trying to subvert defense products by sneaking trojan horses into the Linux source base. If he does have such evidence, he did not mention it.
Overall, I think his argument against open source is pretty weak. I don't think defense agencies should give it much heed.