Why is HTTPS not the default protocol?
SSL/TLS has a slight overhead. When Google switched Gmail to HTTPS (from an optional feature to the default setting), they found out that CPU overhead was about +1%, and network overhead +2%; see this text for details. However, this is for Gmail, which consists of private, dynamic, non-shared data, and hosted on Google's systems, which are accessible from everywhere with very low latency. The main effects of HTTPS, as compared to HTTP, are:
Connection initiation requires some extra network roundtrips. Since such connections are "kept alive" and reused whenever possible, this extra latency is negligible when a given site is used with repeated interactions (as is typical with Gmail); systems which serve mostly static contents may find the network overhead to be non-negligible.
Proxy servers cannot cache pages served with HTTPS (since they do not even see those pages). There again, there is nothing static to cache with Gmail, but this is a very specific context. ISPs are extremely fond of caching since network bandwidth is their lifeforce.
HTTPS is HTTP within SSL/TLS. During the TLS handshake, the server shows its certificate, which must designate the intended server name -- and this occurs before the HTTP request itself is sent to the server. This prevents virtual hosting, unless a TLS extension known as Server Name Indication is used; this requires support from the client. In particular, Internet Explorer does not support Server Name Indication on Windows XP (IE 7.0 and later support it, but only on Vista and Win7). Given the current market share of desktop systems using WinXP, one cannot assume that "everybody" supports Server Name Indication. Instead, HTTPS servers must use one IP per server name; the current status of IPv6 deployment and IPv4 address shortage make this a problem.
HTTPS is "more secure" than HTTP in the following sense: the data is authenticated as coming from a named server, and the transfer is confidential with regards to whoever may eavesdrop on the line. This is a security model which does not make sense in many situations: for instance, when you look at a video from Youtube, you do not really care about whether the video really comes from youtube.com or from some hacker who (courteously) sends you the video you wish to see; and that video is public data anyway, so confidentiality is of low relevance here. Also, authentication is only done relatively to the server's certificate, which comes from a Certification Authority that the client browser knows of. Certificates are not free, since the point of certificates is that they involve physical identification of the certificate owner by the CA (I am not telling that commercial CA price their certificates fairly; but even the fairest of CA, operated by the Buddha himself, would still have to charge a fee for a certificate). Commercial CA would just love HTTPS to be "the default". Moreover, it is not clear whether the PKI model embodied by the X.509 certificates is really what is needed "by default" for the Internet at large (in particular when it comes to relationships between certificates and the DNS -- some argue that a server certificate should be issued by the registrar when the domain is created).
In many enterprise networks, HTTPS means that the data cannot be seen by eavesdroppers, and that category includes all kinds of content filters and antivirus software. Making HTTPS the default would make many system administrators very unhappy.
All of these are reasons why HTTPS is not necessarily a good idea as default protocol for the Web. However, they are not the reason why HTTPS is not, currently, the default protocol for the Web; HTTPS is not the default simply because HTTP was there first.
While there are great answers already given, I believe that one aspect is overlooked so far.
Here it is: Plain HTTP is the default protocol for the web because the majority of information on the web doesn't need security.
I don't mean to belittle the question, or the security concerns of some web sites/applications. But we can at times forget how much web traffic:
- contains only completely public information
- or has little or no value
- or where having more visitors is seen as increasing the value of the site (news media, network effect sites)
A few quick examples, I'm sure you can quickly make more in your mind:
- Almost all company websites, sometimes called "brochure-ware sites", listing public information about a company.
- Almost all of the news media, blogs, TV stations, etc that have chosen advertisement support as their primary monetization strategy.
- Services which may offer logins and additional personalization, but who also give away their content for free to anyone browsing anonymously (YouTube fx).
- It puts significantly more CPU load on the server, especially for static content.
- It's harder to debug with packet captures
- It doesn't support name-based virtual servers