So, after waking up early and working from home until after lunch, I had to ruin my productivity and return to a roughly zero average by going to the office in the afternoon and allowing entropy to take over - not that I had a chance, but I really ought to have stayed home.
Google Web Accelerator
Anyway, as a sideline of some testing I did over my home UMTS link, you'll find a small set of notes on Google Web Accelerator.
That thing is scary - not due to the privacy issues (I'll let the pundits go into those), but due to the sheer power involved.
Having a humungous network of machines acting as proxy is no news to me (I use Coral to divert excess traffic from my site), but having Google do it for you with this neat level of browser integration is something else.
I think the likes of ByteMobile should be getting worried. Even considering that Google does not do image and content re-compression (no, Google uses plain old - and standard - gzip and HTTP 1.1 mechanisms on the client side), it has effectively nullified some of their basic product features.
That said, although both my home and office connections were fast enough for the compression to make nearly no difference (I always configure Firefox to do pipelining anyway, and most of the sites I visit use gzip compression), pre-fetching is a different story.
Update: even though I added a reference to my notes, everyone keeps mentioning current problems with pre-fetching, so I thought it best to add the link here as well. In short, pre-fetching would cause less problems with web applications if it stuck to links marked as such and avoided JavaScript links.
It was the one feature that really sped things up for me, and I expect it to start cropping up here and there, if only as a cue from webmasters as to what links might be important to follow next:
<link rel="prefetch" href="http://url/to/get/">
Now all we need is a LAN version for home and office use, so that non-Windows machines can benefit from all this goodness (I personally hate to install this sort of thing on individual machines - cached content is meant to be shared, not hoarded).
If you're OK with the (so far mostly hyped) privacy issues, of course.