Some photofinish/finishlynx timers are good enough to get competely accurate results posted within 1-2 seconds of a runner finishing, which will almost be instanteous, if you're watching.
However, some timing systems have photoeyes/beams incorporated which can stop the clock at a certain time. Depending on the placement of the beam, this may bea relatively reliable way of stopping the clock.
Even further, FinishLynx can offset this time, so that it can add additional time to the time when the beam breaks, in order to not display a time faster than what was actually run. You can set this time to whatever you want to, but .02-.04 seconds is rather common.
Finishlynx can also incorporate chip systems, which are more frequently being built into tracks. Basically, the chips can be used for unofficial splits, but their formal time is still based on their torso.
Lynx developed something called IsoLynx that has a similar function, it debuted a year or two ago.
Finally, the reason chips are used at big road races? The distance is often fairly arbitrary, and formal results don't matter as much. It's a lot easier to time a race of 1000 people with chips than torsos...
In NCAA meets with chips, a digital video system has to be used in conjunction, in order to break "ties" where athlete finish within .4 of a second of one another (by chip time). Different states have different rules about this, some of which are pretty daft. There was a case in Alaska a few seasons ago where a kids foot beat the other kids foot to the mat, and registered a faster time, but their torso did not. The kid who technically won was then second because their state rules were poorly written.