Widgets are web pages embedded in larger web pages, generally using iFrames — the content comes via a separate HTTP connection and has its own CSS stylesheet, cookies, etc. Final composition takes place in the user’s browser.
Portlets are software modules that produce fragments of HTML markup that are assembled into a single HTML page, sharing common CSS stylesheet, cookies, etc. Final composition takes place on a portal server, and a single page is delivered to the client browser.
Portlets have a lot of features that iFrames don’t: they require fewer HTTP connections, they allow for common styling (one CSS stylesheet can style all the portlets on a page), and they can communicate with each other and take advantage of common authentication/authorization, etc. (so that a user doesn’t have to sign on to each portlet separately).
Portlets use a window-manager metaphor, allowing the portlet server to resize them, expand them etc. They also have modes, like edit and view, all of which can be accessed through a common interface. All of this happens on the server side.
iFrame-based widgets don’t normally do any of that, but they don’t require special portal servers, they can be embedded in more creative ways, and they offload the processing from the server to the client. They also introduce potential security holes, but only if they’re hosted somewhere that’s not under the original company’s control (the same applies to remote portlets using WSRP).
Portlets are used mainly in intranets, to provide a collection of enterprise apps on a single web page for employees (e.g. a news feed, calendar, expense forms, bug reports, etc.).
Widgets are used everywhere else (e.g. embedding Google maps, Facebook applications, etc.). While widget authors/consumers don’t tend to know (or care) much about portlets, the portlet people haven’t failed to notice the popularity of widgets — most (if not all) portal servers now have an iFrame portlet that does little more than wrap an iFrame and allow it to be resized, etc.
Are the extra features of portlets compelling enough to justify the extra cost and hassle of running a portlet server? Now that we have browser tabs, AJAX, etc., do enterprises really need to continue to squish all their apps into a single web page that looks like a 1995 Mac desktop gone bad?
My guess is that the only portlet feature with compelling benefits is common authentication/authorization — once the web community gets behind a solution to that problem (OpenID or something similar), widgets will probably push portlets out completely, even in the enterprise.