This suggestion ties together with "Grant scripts in browsers access to skin data."
Browsers already provide a means of processing data on the client. I feel much potential has come from the ability of the server to execute code in a client's browser, but why stop there?
I propose that macros, buttons, inputs, and any other control that can execute verbs be given a means to execute functions in a browser. The script may then interpret the input, adjust it, throttle it, send instructions to the server, or perform some magic as proposed in the suggestion linked above.
The major benefit in this is having lagless interface controls. Currently, in order for a button to intelligently modify the interface, (such as appending text to an input control, or performing an action based on the state of other controls) it must execute a verb on the server, which must query the client's controls with winget, and then apply changes with winset. A script in a browser can perform such complex manipulations of the interface instantly, and with no hit on the server.
Bonus points if the browser could hook into and listen for events such as mouse up/down, and resizing.