The FINA olympic size is precisely defined as 50.0m with a tolerance of plus/minus 1cm, ie 1/5000th of the size of the pool (*). Since it takes about 20s to cross 50m, the variation in time taken is on the order of 20s/5000 = 0.004s, less than half a hundredth of a second.

The crystals in your computer or mine measure billionths of a second reliably, and separating swimmers to within four thousandths of a second with literally negligible margins of error is easy with touch panels. There's really no issue at all of margin of error in time measurement to speak of.

When athletes are selected, it's safe to say that they are believed to be able to reliably repeat their performances. It would not do to send an athlete whose performance has a huge random variation from race to race, and since the olympic performers are practically constantly monitored during training, such variation does show up.

I have little doubt that a 0.01s difference is meaningful on the day of the race.

(*) I believe Beijing was boasting a tolerance of 2mm rather than 1cm, but if you're comparing against other olympic races in other countries, 1cm is needed.

--
$E(X_t|F_s) = X_s,\quad t > s$