Real Time Costs

The ratio of aggregate times spent using two different interface designs for the same task is probably our best measure of relative efficiency, but Tognazzini's observation implies that user perceptions aren't a good measure of either. Accordingly, an entire discipline of quantitative interface analysis has grown up around doing actual timings. [Raskin] gives examples and discusses several efficiency methods. Here are some averaged timings, combined from [Raskin] and [Lewis&Rieman]:

Table 4.1. Timings for various interface actions

Enter one keystroke on a standard keyboard:.28 second Ranges from .07 second for highly skilled typists doing transcription, to .2 second for an average 60-wpm typist, to over 1 second for a bad typist. Random sequences, formulas, and commands take longer than plain text.
Use mouse to point at object on screen1.5 second May be slightly lower — but still at least 1 second — for a small screen and a menu. Increases with larger screens, smaller objects.
Move hand to pointing device or function key.3 second Ranges from .21 second for cursor keys to .36 second for a mouse.
Respond to a brief light.1 second Varies with intensity, from .05 second for a bright light to .2 second for a dim one.
Recognize a 6-letter word.34 second 
Move eyes to new location on screen (saccade).23 second 
Decay time of visual perceptionAverage of 0.2 second, range from 0.09 seconds to 1 second. 
Decay time of auditory perceptionAverage of 1.5 second, range from 0.9 seconds to 3.5 seconds. 
Retain an on-screen message in short-term memory10 seconds 
Retrieve a simple item from long-term memory1.2 second A typical item might be a command abbreviation ("dir"). Time is roughly halved if the same item needs to be retrieved again immediately.
Learn a single "step" in a procedure25 seconds May be less under some circumstances, but most research shows 10 to 15 seconds as a minimum. None of these figures include the time needed to get started in a training situation.
Execute a mental "step".075 second Ranges from .05 to .1 second, depending on what kind of mental step is being performed.
Choose among methods1.2 second Ranges from .06 to at least 1.8 seconds, depending on complexity of factors influencing the decision.

These timings have many interesting implications just as stated. Here are a few:

Well-designed interfaces only support one method for each task.

We get this from the high time cost of choosing among methods. Accordingly, it's best to avoid duplicative controls or commands. A partial exception to this is if a GUI has duplicated controls on two screens or panels that never appear at the same time; in that case, the user never has to choose between them.

Well-designed interfaces avoid overusing the mouse.

Mouse selections are expensive. These times also should serve as particular cautions to designers who build elaborate forms-based interfaces that involve alternations of mouse actions and typing.

Well-designed interfaces don't assume that the user will retain information once it has vanished offscreen.

Raskin makes this point about error messages, but it applies to other kinds of information as well. Unless a user is somehow prompted or motivated to put information in long-term memory, the details of on-screen data will fade in about ten seconds, even if the general sense remains. This is a powerful motivation for the Rule of Transparancy: Every bit of program state that the user has to reason about should be manifest in the interface.