First, let's talk Border Gateway Protocol. The internet is composed of thousands of endpoints known as ASes (Autonomous Systems), and they route data with a protocol known as BGP (Border Gateway Protocol). In recent years the size of the BGP routing table has been exponentially increasing in size, breaking well over one 100,000 entries. Even with routing hardware increasing in power, it is barely able to keep the pace with the ever-expanding size of the BGP routing table.
The tricky part in our MITM scenario is that BGP implicitly trusts routes that other autonomous systems provide it, which means that, with enough spamming from an AS, any route can lead to any autonomous system. It is the most obvious way to MITM traffic, and it's not just theoretical - Defcon security convention's site was redirected to a security researcher's website in 2007 to demonstrate the attack. Youtube was down in several Asian countries when Pakistan censored the site and mistakenly declared its own (dead) route the best for several ASes outside of Pakistan.
A handful of academic groups collect BGP routing information from cooperating ASes to monitor BGP updates that change traffic paths. But without context, it can be difficult to distinguish a legitimate change from a malicious hijacking. Traffic paths change all the time to cope with natural disasters, company mergers, etc.
Next to discuss on the 'Global MITM attack vectors' list is Domain Name System (DNS).
Although ISC's Fine DNS server BIND has stood the test of time and come out relatively unscathed (as have Microsoft and Cisco's DNS offerings), a few notable vulnerabilities have been found that could potentially jeopardize all traffic using canonicalized names on the internet (i.e. practically all traffic).
I won't even bother discussing Dan Kaminsky's research into the DNS cache poisoning attack, as it has been beaten to death elsewhere, only to be awarded 'most overhyped bug ever' by Blackhat - Las Vegas. However, several other DNS bugs exist that have severely compromised internet security.
The Dynamic Update Zone Bug crashed DNS servers and had the potential to remotely compromise machines and DNS caches.
The Transaction Signatures Bug allowed for full remote root compromise of any server running BIND at the time the vulnerability was announced, obviously allowing DNS entries to be compromised.
Finally, we must discuss ARP Poisoning, 802.11q Retracing, STP-Trunk Hijacking, RIPv1 routing information injection and the slew of attacks for OSPF networks.
These attacks are the 'familiars' to a network admin for an independent company (rightfully so, considering these may be the only ones they have control over). Discussing the technical details of each of these attacks is slightly boring at this stage, as everyone who is familiar with basic information security or TCP has learned ARP Poisoning. The other attacks are likely a familiar face to many network admins or server security aficionados. If these are your concern, there are plenty of very good network defense utilities that exist, ranging from Free and Open Source utilities like Snort to the enterprise level software from Cisco and HP. Alternatively, many informative books cover these topics, too numerous to discuss, but several I've found helpful in the pursuit of network security include The Tao of Network Security Monitoring, Network Security Architectures, and the classic Network Warrior
In any case, I find it somewhat disturbing that people assume that these sort of attacks require ISP or Government level access. They require no more than the average CCIE has in networking knowledge and the appropriate tools (i.e. HPING and Netcat, not exactly theoretical tools). Stay vigilant if you want to stay secure.