Feedback about the performance comparison between Curl and ActionScript has focused more on the importance of JPEG encoding than on the importance of high performance in a rich client language.
There have been a number of comments indicating that JPEG encoding is indeed useful in some rich client applications, enough to prompt discussion about whether it could be built into the FlashPlayer (where, presumably, it would not be coded in ActionScript).
However, image processing is just one example of an application genre where substantial application specific processing is important. Others include business intelligence, financial analysis, simulation, and product configuration. Speculation about the potential and implications of high performance language alternatives is beginning.
When considering performance, its important to distinguish between the performance of components that are built into the platform, and the performance of components developed specifically for one application. For example, the guimark test is intended to exercise the platform, not its language. From the theory page: "Code execution should have a negligible impact. The user code in each of the tests has almost zero impact on the test case"
We chose the JPEG encoding task to focus on the opposite end of the spectrum: exercising user code. The JPEG task is well suited for comparing language performance, because it has very little reliance on library functions, which might well be written in a different language. Furthermore, the operations used doing JPEG encoding (iteration, floating point math, memory allocation, object creation, method and function calls) are used in a wide range of applications.
Of course, every application uses a different combination of language and platform features, so generalization from any particular test to a specific use case must be done carefully. RIA language performance is most relevant for applications where a significant part of the application logic is compute intensive.
The distribution of processing between tiers will vary according to the nature of the application, but one wants such decisions to be based on application characteristics (data flow and work flow) rather than on intrinsic limitations of the client platform.