Analysis done on May 7th 2014 and updated on May 13th.
While some of our API calls like imageinfo follow the general pattern of higher usage on the weekend, the general API requests per second graph shows us that there is no correlation, i.e. our API calls aren't creating the big spikes seen on global API count because there are days where it's clear that imageinfo is going down and global API calls stay up.
Data suggests that Media Viewer's impact can't be seen, as Media Viewer's peak of thumbnail misses was between April 18th and April 21st, at what point the Image Scalers graph is actually lower than usual. Image scalers graphs seem to have been consistent through out latest large launch (French + Dutch).
Swift sees a slight increase over the period of the French/Duthc launch, but in previous weeks there were much higher peaks. However it seems like there is a strong correlation here between the peaks and the times when the Image Viewer thumbnail Varnish misses where at their highest. The good news is that the amount of misses has since then stabilized and seem unaffected by the most recent pilot launch.
My theory is that a lot of content comes from Commons and the best Commons content is likely to be used on several language wikis. Thus, but the time we get to bigger wikis, the smaller wikis have already taken care of the thumbnail generation for the majority of the best Commons content. I think that the gradual rollout is working in our favor.
There's a peak over the last few days, but the previous week saw peaks that were was just as high. And those don't correlate with the amount of thumbnail misses seen in Media Viewer.