Run a test for yourself using the settings:
setCookie https://en.m.wikipedia.org disableImages=1 navigate https://en.m.wikipedia.org/wiki/Barack_Obama
Update 22nd December
Originally T110615 suggested removing images would halve first paint for 2G connections.
However, I've booted up a new test to get some fresh results
http://www.webpagetest.org/result/151222_FS_15H3/ vs http://www.webpagetest.org/result/151222_Y3_15EH/ which shows similar results.
I enabled the cookie with values 0 on 1 so that they are subject to similar caching rules.
With images disabled via cookie:
First paint: 68.973s
Bytes: 385 KB
With images enabled via cookie:
First paint: 63.379s
Bytes: 970 KB
On 2G the Barack Obama page hits first paint at 46.131s with images and 19.794s without images . In the former the page doesn't fully load, however in the latter it does (albeit sans images). This is pretty huge.
Daring proposal: We currently allow the disabling of images via the enwikidisableImages cookie - why not do it for all page views. (hear me out...)
A more complete solution would only do this for images beyond the lead section (given that the first image is likely to be the only one the user needs to see to feel like the page is responsive).
Given the images would then be loaded after the startup module, things could get ridiculously more fast.
Proposal: Let's try this out and test it and report back findings.
Let's avoid a debate about whether this is considered breaking the web at this point - let's just work out what it buys us.
- With a flag enabled e.g query string parameter AND disableImages cookie enabled any disabled images are pulled in via JS (when scrolled to - see getBoundingClientRect / debouncing)
We'll get this deployed and run some tests across a collection of pages in both modes and report back what we find