Basically the whole post can be summed up via these choice quotes:
Overall, I disagree with his title, but I don’t disagree with his results ... his methodology, it’s mostly fine
Guy tested partial SPDY page loads, not full SPDY page loads. More specifically, he tested this case: if you upgrade your primary domain, but few of your other domains, your CDN, etc, how does SPDY perform?
(followed up with a table showing that this is a very realistic test scenario)
Well, not entirely. He also explained that the original test counted sub-domains that the primary page owner also controls as non-upgraded third-party ones, whereas someone performing a SPDY roll-out would most likely also upgrade them. The table then showed exactly how many resources that were counted as "non-upgradeable" were in fact most likely upgradeable (most or even almost all of them).
Overall, I disagree with his title, but I don’t disagree with his results ... his methodology, it’s mostly fine
Guy tested partial SPDY page loads, not full SPDY page loads. More specifically, he tested this case: if you upgrade your primary domain, but few of your other domains, your CDN, etc, how does SPDY perform?
(followed up with a table showing that this is a very realistic test scenario)