Why are you measuring web speed in computer time instead of human time? It doesn't matter if it takes 200ms or 170ms to render the page. What matters is how long it takes the person visiting the page to complete the task or obtain the information. Going to gmail, clicking on a subject line and having Ajax load the content is much faster for me the reader, even if the browser could have loaded a plain HTML page faster.
"The more devices there are out in the population, the more enticing it is for developers to develop for them."
Wrong wrong wrong.
I don't want to develop my software and have to test it on 30 different phones. I want to develop for a single phone that has gigantic market share.
It's a phone remember. Getting good performance from a device that necessarily has quite limited CPU and GPU power means that you will need a lot of optimizations to ensure quality. I don't want to have test on a big pile of devices, and I don't want bad reviews because my software performed poorly on one particular device.
As a consumer, yes I want choice. As a developer, it's much easier to make good software for a single OS + hardware combination.
I'm sure the GP is referring to the Spanish Job Study, which has been debunked by everyone who has read it. This article is particularly good at pointing out the massive methodological flaws in the study.
Nothing succeeds like excess. -- Oscar Wilde