Very late answer, but well, maybe it will help someone else who reads this.
halted halfway through
2D experiments are acquired as a series of increments where $t_1$ is changed in order to generate the indirect frequency dimension ($F_1$).
On Bruker spectrometers, 2D pulse programmes are always written so that each increment is saved to disk as soon as it has been recorded. So, if you were running an experiment with 256 increments and you stop acquisition halfway you will still have 128 increments saved to disk; there is no need to use the
tr command. Furthermore, since $t_1$ increments are usually acquired in a regular, uniform manner, this truncated data will still be suitable for Fourier transformation.* This is already pointed out in Mad Scientist's answer.
However, it omits one salient point: if you record only half of the increments for a 2D experiment, then the resolution in the indirect dimension will be decreased by a corresponding amount. Depending on the processing being used you may also find that there are truncation artefacts, which appear as 'wiggles' around the peak in the indirect dimension (although processing parameters are often chosen to minimise these anyway).
Whether that is 'usable' enough will vary depending on your use case. If you are looking for NOEs between two peaks that are quite close to one another, it's possible that it's not good enough. On the other hand, HSQC spectra are very sparse and you can often get away with it.
If you're unsure, you can test this out by processing the 2D data during acquisition, i.e. before stopping it. Just do
xfb (Fourier transform), and if necessary
apk2d (phase correction), plus whatever other processing you would normally do. TopSpin will process the data which have been acquired so far. If the results are good enough, you can stop the experiment. If they aren't good enough, don't stop it.
* Off-hand, I'm not sure what happens with NUS spectra, but I expect that it can still be processed as usual (you'll just end up with a lower NUS sampling percentage than what you initially keyed in). Note that using a very low NUS sampling percentage can—and will—introduce reconstruction artefacts.