This seems to be comparing data URIs to an essentially un-optimized page. It would be more valid, I think, to compare the load time for different levels of optimizations which are more browser-compatible: e.g. CSS sprites, using different image domains, etc.
That said, I understand that any browser/HTTP performance testing is kind of a pain, so thanks for this result.
I agree that a comprehensive test would include the optimized scenarios that you mentioned. In this case the intent was to show that even when comparing two extremes - a very unoptimized page and a super-optimized page - the results were not as disparate as we'd expect. Scenarios involving sprites and domain sharding we'd probably expect to fall somewhere in the middle. Thanks for your comments.
There is no methodology here as to whether the test was local (~0 latency) or simulated-real (50-200ms roundtrip latency). The latter amendment might change the picture substantially.
Hi Jeremy, I'm the author of the post. The tests were not conducted locally, but over a real network, albeit a low-latency one. Latency is definitely a key factor, so I'm planning to run some more tests under different network conditions. Thanks for your feedback.
I had a suspition that there wouldn't be a huge benefit. HTTP has pipelining for a reason.