In enterprise software, the cost of the product is usually a tiny part of the TCO, the bulk of which is typically labor (whether in-house or consulting). In some cases, it is far more important to buy a product that is popular (and thus comes with its own labor pool from which to recruit) than to save money by finding a less-expensive product. In other cases, a particular integration product that connects your particular endpoints may be rare enough that no one would ever bother to build it unless they could get reimbursed reasonably.

That's why it's not surprising that there is free a free version of something common like Java middleware (JBoss) that's pretty solid (but not worth the trouble of switching to if you have a significant investment in a proprietary vendor's tools), but if you want a less common tool like one that allows you to plug an asp.net web application into a Java portal framework, you should expect to pay a premium for it.

I think the reason this hasn't been addressed from a startup standpoint is that these lessons have been absorbed so thoroughly that only startups with a really clear, well-defined niche would even consider the possibility of going into the desktop software market, because everyone knows it's a tough row to hoe. Web business models like advertising are premium accounts a lot more plausible, even as it look like web advertising may be looking at a correction.