I don't need to think anything about Binney's motives to reason about the possible scope of surveillance on IP traffic.
Based on what you'd have to do to capture foreign traffic as opposed to US traffic it makes sense that the Room 641A sites are intended to capture foreign traffic. All the public information supports this. There isn't any public disclosure of the type of infrastructure needed to comprehensively monitor US IP traffic.
If they are trying to capture foreign IP traffic there would need to be a intercept device where the foreign network interconnects with the US networks and where cables land. Large foreign providers tend to have 20 or fewer IP POPs in the US clustered in the same places. Look at networks maps from Tata, ReTN, NTT, etc. You will see many of the same locations. To tap them you'd put a intercept site in Seattle, Bay Area, Chicago, Miami, New York, and Virginia. Maybe Atlanta & Texas.
By the nature of the internet US traffic would get caught in that. E.g. ReTN had too much inbound traffic so they sold a college in Chicago some cheap outbound traffic. Now ReTN loses a router and their peering goes down so the bits to that college get hauled out to Europe and back. Bits just got captured. ThinThread (Binney's program) reportedly accounted for that and would mask US person PII pending a court order to decrypt it.
The fact that Trailblazer doesn't mask PII is a issue, but both programs would have captured US person PII, the difference is the addition of a bureaucratic hoop that the current and past administration would order jumped in a heartbeat. It isn't like ThinThread required a FISA Judge to walk over with a crypto token and descramble the data.
The nature of the problem means some US traffic will be caught but that does not make it a domestic spying program.
Unlike many other countries with a domestic internet surveillance regime (Syria, Egypt, Iran just to name a few) the US does not mandate that traffic pass through the national carrier or otherwise limit connections within its borders. Syria can drop in a few Blue Coat boxes and catch everything. In the US you'd need thousands, if not tens of thousands of intercept devices -- one or more for every edge in the graph of ASs (http://www.caida.org/research/topology/as_core_network/). It is not a insurmountable problem but it would take a act of Congress (likely challenged to the Supreme Court) and the cooperation of a lot of people to implement. It would not happen overnight or in secret.
Now, US providers are required to support CALEA but that is targeted at individual subscribers and most manufacturer implementations aren't suited to wholesale spying.
This isn't to say 641A sites couldn't work in hand with CALEA, just that it would be a bit more legit than "omg nsa taps all the wires". E.g. 641A could be watching traffic to/from a Jihadi site hosted in France without a court order under the Protect America Act of 2007. NSA notes someone from a US IP has been posting things that indicate radicalization. NSA passes it to FBI to check out. FBI gets a court order for a CALEA tap for the US IP. FBI sends order for tap to ISP.
No wholesale tapping of US traffic is necessary. Of course in this example watching anything to/from the US without a FISA order is likely not kosher prior to PAA hence the warrantless surveillance controversy.
When fearmongering over home-grown lone wolf terrorism get louder, that's when the real trouble starts (except they probably have all the info needed for that over the counter from TransUnion etc).