Murta, thank you for the suggestion. I do make heavy use of ArrayFormulas in some of my plug-ins. I do that mainly through .NET frameworks, and that works very well (since I don't have the advantage of short text programming, as we have been used to with Mathematica, recovering pieces of code from different languages becomes very useful). ArrayFormulas work very well, although sometimes users get confused with all the cells blocking and the ctrl+shift thing. It is also through that same framework that I have called the cloud, just for some tests (the x^2 function).
Ended up discovering that the only thing that I will be able to do with it, will be a licensing control system through the cloud. Nowadays, I have my packages blocked to specific computers (serial number stuff), and when I lend the packages to a partner, or a client, I create a USB dongle with limit count of uses (just enough to share some work, during a specific project). But this forces me to manage a physical object (send it by mail, recover it at its end of life, etc). Since work sharing is not as mission critical as production, what I can do through the cloud is activate or not the plug-ins, at their first call of each session, for whichever software they were developed. That way, computation is still made locally (though, and shamefully, on .NET), but activation is controlled at a distance.
On my line of business (and I believe, it is similar to most of the engineering world), my computations need to be heavily customizable, to fit each application needs. This means that my functions are just small bricks that are putted together by the end user for a much bigger analysis. This means that the portion that can take use of arrays is very small (10 to 15%?). All the rest requires individual calls.
I have never tried webMathematica, since I really don't feel that it is fitted for my applications. As explained before, there are a lot of times that the engineer is abroad, and sometimes even without access to the internet (on the client site, etc), and trying to receive thousands of answers from a distant server, just because he changes something on a CAD or a spreadsheet, is out of the question. Since I only have time to invest my-self on one technology, I have to chose the one that answers to my needs in the best away. And unfortunately, not due to technical limitations, but to business strategy (in my personal opinion, deviated of what are the needs of millions of potential users, since I don't believe that WR really intends for cases like mine to pay for a full Mathematica license for each end user; if that was the case, webMathematica would also not allow for such solutions), WR technology doesn't present itself as a feasible option. And still, everything is there, and I can't avoid thinking that this new player version is an opportunity that was lost (as the one before, and the one before that one, etc), but this time, most specially with the fusion that was done with the Player pro, the availability of the entire wolfram kernel, etc.
But this is only my humble opinion (although shared with many other users I've encountered during the conferences).