Earlier this week, I wrote about a survey that finds Apache falling and Microsoft rising in the Web server race. Not so fast. Another recent survey lends itself to a much muddier picture.Netcraft's August 2007 survey shows Apache servers hosting 48.4% of active sites on the Web, Microsoft hosting 36.2%, and the gap narrowing. However, another survey by E-Soft -- which, admittedly, the Apache guys pointed me toward -- finds the exact opposite trend. In its survey, Apache has 73.81% to Microsoft's falling 19.33% share.

I could say there are lies, damned lies, and statistics, but that would only be half the truth -- pun intended. The methodologies of the two surveys vary significantly, and I can't for the life of me tell which one is more accurate.

Even Microsoft doesn't jump on the Netcraft numbers. "Market share numbers are always more art than science, but the good news is that we seen good traction with IIS and that we've seen a good turnaround," said Bill Staples, Microsoft's product manager for the company's Internet Information Services Web server, pointing to improved security and reliability features in IIS 6.0 that have helped overcome Microsoft's bad Web server rep but not pointing to any verifiable market share numbers.

Netcraft recently began counting Blogger sites, which were formerly categorized in the Apache bucket, under a new Google category. That's caused a hit for Apache, as Netcraft now puts Google at 4.4% of all Web servers. A few bloggers have suggested that this is misleading because Google uses some version of Apache for its servers, but when I contacted Google, I was told that, "in general, Google does not use the Apache server." Still, that makes me suspect of Netcraft's numbers. Why the sudden shift?

Meanwhile, Apache has a bone to pick with Netcraft numbers. "As noted by Netcraft itself, Microsoft actively seeks to migrate domain parking companies like GoDaddy to the Windows platform," Apache developer Sander Temme wrote in a February blog after Netcraft reported similar losses. "Winning over a single domain parking registrar can cause a notable shift in market share, which causes the graphs to spike. Such a shift is absolutely meaningless except Microsoft can use it as a sales tool, and use it to win over potential customers."

OK, but Netcraft says it filters out parked sites for its findings on "active sites." The survey measured almost 128 million overall sites on the Web, but only a little more than 55 million were active. Each domain is only counted once. However, what's not filtered out are "some companies with a very large number of domain names referencing a single IP address," according to a document outlining Netcraft's methodology. That document says there are 5.5 domain names for each IP address, and that figure could swing results heavily in different directions if we're just counting the number of physical servers out there as opposed to who runs the most Web sites.

E-Soft surveys fewer sites, just under 25 million. It, too, avoids parked sites, counts domain names only once, and figures out the type of server running a site from HTTP header information. However, E-Soft goes in the other direction from Netcraft with single IP addresses running numerous sites. "If we find more than a reasonable number of Web sites on one IP address, we're dropping them all out," E-Soft director of technology Tom Reinke told me. E-Soft also doesn't count "small domains that nobody links to." Therefore, it isn't counting all the sites.

It seems we're back to square one. Lies, damned lies, and statistics it is.