I am in the process of evaluating/learning portal and portlet technology for my company. We are a HUGE organization, so there have been several efforts of exploration in various groups within the company. In fact, the "official" company-wide web-app and services framework team has actually implemented portlet functionality into the proprietary framework. I don't like the level to which they have abstracted away everything, though, so I'm making it part of my effort to evaluate their implementation. We use WebSphere Portal Server (we are a huge IBM shop, so it was the only realistic choice).
Thus far, my impressions of Java portal/portlet technology is that it is over-hyped. How many applications really have Yahoo-like requirements? Also, the portlet spec is over 18 months old at this point, but there is no public information about any activity on it. IOW, it appears to be stalled at the 1.0 level, and that 1.0 spec is quite immature. There are several key ideas that the vendors have all solved in proprietary ways that need to be part of the standard API and spec. The authors of the spec are remarkably silent on the progress of a 1.1 or 2.0 version - my educated speculation is that there is little or no progress happening at all.
The user activity I keep seeing (a Yahoo Group, articles, blog entries, etc) seems to be inordinately focused in the offshore population. I wonder if the offshore/outsourcing companies are hearing the hype from management types and trying to find ways to inject portal/portlet into every project they contract now...
Performance also appears to be a major issue. The extra processing required for every single user request is multiple times that for servlet apps, and, especially within our company, that results in apps that can handle significantly fewer concurrent users per unit of hardware than equivalent servlet/JSP apps. IBM claims that latest versions of WPS are significantly better than previous versions, but also seems to acknowledge that it is not equivalent to plain-old servlets and may never beÂ - the stated hardware requirements for running Portal Server are pretty high-end.
Something else: I haven't seen much talk about portal in terms of transactional applications. I can see that having a "main" app as a portlet surrounded by little helper stuff like calculators, mail checkers, stock quotes, etc. would be "nice to have" candy, I haven't seen any examples of highly transactional portlets all coexisting and sharing window space.
Another potential problem is that of screen real estate - you UI designers have a significantly harder job when you're dealing with multiple portlets on every page; it means none of the portlets have as much real estate to work with as they would if they were individual pages or traditional web apps.
I think the portal idea can be very useful for certain applications that have specific requirements about incorporating multiple data views onto one page. I just don't think many applications actually have such needs. When they do, implementing it in plain servlets and JSP can be done, but would require building something like the portlet API and container anyway. It's just that I prefer to build something like that as-needed and specific to what you really need, rather than relying on a huge, bloated, wants-to-do-everything implementation like is provided by IBM, BEA, etc.
As I said, I am still learning and evaluating and planning some PoC mini-projects to explore these topics. Thus, my opinions are young and likely to change at least a little over the next few months