Ok, here are the facts and figures from my experience. I will list what I know and you can arrive at your own conclusions.
My primary site is a large user-created-content site with 50,000+ pages of user articles, plus a 20,000 or so support pages (member information, etc.) These articles cover over 400 categories of topics, so it covers a broad range of subject matter. I record the traffic logs in a database instead of regular server logs so I can more easily parse the data. Previous to the change I was getting about 33,000 referrals from Google per day. Even though I had dedicated servers, my ISP was getting slower and slower. YSlow was showing page load time of about 1.65 seconds on average, with my home page loading at about 3.5 seconds. YSlow shows my biggest problem as too many HTTP requests. I don't have large images on this site, just a lot of smaller optimized images. I tend to add features primarily with the end user in mind, trying to remember general principals of optimization, but most of my pages contain 10 or more database queries and they are not uber-optimized for page load time. My focus has always been on providing cool features and not the fastest page load times. Because of the large number of pages, I have never tracked keyword ranking, it would be a futile waste of time.
I decided to switch to a local Internet Service Provider (a friend) who had just installed a fiber connection. I built a new server for the switch, and upgraded to SQL Server 2005 (from SQL 2000) .
10 days ago (Saturday March 8th) I made the switch. Here's the specs on the hardware:
2.8 Ghz P4 with Hyper-threading (remember those?)
2 Gigs of RAM
10,000 RPM Barracuda SATA drives
SQL Server 2000
Dual Bonded T1's, which are supposed to equal 3 megabits up and down but my tests showed an average of about 0.3 megabits up and down (this was a definate bottleneck).
2.33 Ghz Dual Core
4 Gigs of RAM
1,500 RPM SAS drives
SQL Server 2005
Fiber from my local ISP (Comcast). Rated at 10 Megabits, but my tests show as much as 25 megabits up and down, almost 100 times faster than my previous connection.
Testing shows that SQL Server 2005 processed the queries about 18% faster on the same hardware. On the new hardware the pages are processed about 60% faster.
Page load times were reduced to an average of 0.7 seconds, less than 50% of the previous load times.
Traffic with Google Search referral information increased 11% from the 10 days previous to this switch to the 10 days since this switch.
Page views per visit have increased 9% from the 10 days previous to this switch to the 10 days since this switch.
Without giving specifics (AdSense TOS) the CPM for this site increased 6% from the 10 days previous to this switch to the 10 days since this switch.
CTR hasn't changed. (Don't know why this is.)
This has all equaled a total increase in AdSense earnings for this particular web site of about 26%.
I should say that even though it's only been 10 days since the switch, because of the large size of the site and the broad array of topics, traffic to this site and AdSense CPM has usually been very stable. I expect this 'trend' from the past 10 days is here to stay.
The most surprising thing to me is that the difference was apparent immediately. It didn't take days for traffic from Google Search to increase. It didn't take any time at all for eCPM to increase. It happened pretty much immediately.
For obvious reasons, we have been very pleased with this change. I am the primary programmer for this site and I hate the thought of tediously going through the site, optimizing one page at a time. This is one very rare example of when simply throwing money at a problem actually fixed the issue.