Hi Andrés,
There isn’t a “best practice” in this situation, it comes down to what is the easiest for you and your team.
Programmer sounds would need keys to be passed in, so in code you would need an array of all the necessary dialogue FX keys which you can pull a random value from.
By comparison, you can use one or multiple events with randomized multi instruments to do the same thing. This is the simplest approach and most flexible as you can quickly add more sounds without needing to adjust code or keys.txt files.
In terms of performance, there is nearly no difference. For the case of these dialogue FX, I would suggest using multi instruments as they require the least work to update.
Thanks,
Richard