Without telling us the size of the data set that anecdote is meaningless. Other algorithms are _asymptotically_ better than bubble sort, i.e., when you get a certain amount of data, other algorithms perform significantly better. So this choice matters when the input can be arbitrarily large.

And it's false to say everything's disk/network bound these days. There will always be important data structures and other code that needs to scale, where the right algorithm makes the difference.