I guess that's my point, in using the profiler, I wouldn't expect a handler to take 300ms on a 2.16GHz Core 2 Duo using Firefox. Would you? This is what I'm looking for. For a set of architectural elements, what is the expected performance on a given platform?
No one can answer that question for you, keep in mind that usualy the developer's computer is faster than the contractor's computer.
You must not focus on the "ms" but you must focus on the "%". For example a possible solution for my problem was to make my own tree, which showed to the user only the information needed and did not compute the information not visible for the user. Another solution was not to instantiate the node components, but to reuse them, because when validating the display of a component it make a lot of calculus. This all looked at the scrollbars position.