That's quite interesting, Gene. I have exactly the reverse experience W3 Total Cache gives me something around 4 times less page download time and it uses the browser cache far, far more efficiently (aye, I still use WP Super Cache on many of my low-traffic blogs). So I don't rely on host-tracker.com exclusively; the power of W3 Total Cache is that it's not just about the server-side, but also the way it integrates so neatly on the browser side. Most of the non-textual elements remain neatly on the browser and there is even no communication between the browser and Apache (and, of course, the textual elements are completely static, so they load faster too):
On a dedicated server, there is a whole magnitude of further improvements that you can add! Just activating things like Advanced PHP Cache (APC) or Xcache will make your eyes water with the speed improvements. Go with ngnix instead of Apache, tweak it properly with all kinds of PHP accelerating tricks and tips, make sure that W3TC is happily interconnected with them, subscribe to CloudFlare, and you'll get an old Pentium II with a few hundreds of MBytes at home outperform DH's servers
DH's shared hosting setup is suboptimal for W3TC but it still makes a huge difference; I have no idea what host-tracker.com is actually measuring, but I'm pretty sure it can only measure what happens on the server-side of things. But it takes two to tango: it's how things are improved on the browser side that will make things faster as well, too. I particularly like the way W3TC is very aggressive in compacting CSS, JS, and HTML, and optimizes the order things are sent back to the client; using Google's Page Speed Report I can also see what the culprits are and fine-tune performance based on that — on W3TC, not necessarily on the theme itself.
Sure, it is a messy plugin to delete. It's not "nonsense" in that regard; I prefer the word "messy" And their creators warn in advance that it's not a plugin for everybody; a badly-configured W3TC installation will be worse than having no caches set up...[hr]
Ok, hmm... I just did a test with host-tracker.com. Apparently the free service just tells "Response time", which i don't know what it means. Are you using the paid service, which provides a wealth of further information?
Because "response time" might mean different things. Is it the time of completion for the first request to the site? Then part of that time is just DNS.
The second test was made from a "test" blog which has little traffic, the Twenty Eleven theme, and few active plugins. It's a very simple blog, so I just have WP Super Cache active on it — I never bothered to go through any pains of configuring W3TC on it. It has next-to-zero visitors. Average response time is 3.15 sec and bandwidth 8.78 Kbps (in this case, mostly because there is little to transfer except a few scattered images).
As a reference... I've got access to a virtual server which is fully-dedicated to a WordPress install. This allows me a lot of tweaking on the server side, way beyond what DreamHost allows, even though I have not the full range of options, and it's not a dedicated server (although I believe that the remaining servers have little traffic, since most are not even public, but just used internally for the client). With way more configuration options than DH provides with their own Apache/PHP configuration, and with twice the traffic (on average) than my first example, with W3TC active I get an average of 0.59 sec and 30.41 Kbps (this site has a way simpler interface, but it's not a "bare-bones" site, but a real, live site for a governmental institution, with plenty of images; it just has far less external functionality and a way cleaner theme).
While obviously I don't know what host-tracker.com is actually measuring in the free service, it's a bit pointless to reach some conclusions. What it looks like to me is that W3TC + CloudFlare gives me 3 times the performance (at least) than a very simple, unused blog hosted on DH using merely W3 Super Cache, which is consistent with my other tests; in fact, I get often more than "3 times the performance" because the browser cache makes so much difference.
Nothing beats a virtual/dedicated server with a simple theme, though. The main reason is that it can be optimised at will. If I wish further performance improvements on that client's site hosted on a virtual server, I have way more options, all of them on "stand-by" if the traffic justifies the trouble:
- Add CloudFlare (easiest to setup)
- Add Squid and put it in front of Apache (Squid is faster serving static pages, and W3TC will make pretty much everything static anyway). Note that Squid has gazillions of fine-tuning options, so just that will allow a lot of extra performance improvements
- Use Advanced PHP Cache instead of Xcache on PHP (allegedly it gives better performance; I've never tried it)
- Dump Apache and use ngnix for ultrafast performance
- Start tweaking the Linux kernel
On a very heavy-traffic (old) server with hundreds of sites, a WordPress installation with W3TC, running inside a FreeBSD jail with Squid in front of it — a server that has been heavily optimised with all tricks of the trade — I get pretty much the same numbers as on the virtual server used to serve a single site: 0.51 sec and 21.61 Kbps (the blog has simple themes)
But if I don't have the option to tweak the server — there is a limit to what one can do with DH's shared hosting service — W3TC (and optionally CloudFlare) gives a reasonable compromise. Yes, it's three times slower (assuming "response time" measures slowness...) than a dedicated solution (which, however, are far more expensive than three times the cost of DH's shared hosting services...). But it's also three times faster than using WP Super Cache. I understand that WP Hypercache gives better performance than WP Super Cache, but I haven't tried it out. And, again, that's not counting the user experience with W3TC's aggressive use of browser caches; some pages load so quickly that it hurts the eyes watching them to render... and this cannot be measured easily by external measuring tools.