If you are to do any work on large datasets you need to learn how to handle those efficiently.
Just think if you are going to do image analysis on a 2 Gb TIFF image. Or sequence matching on a 150 Gb genome dataset.
When you move from a "nothing is slow" to a "this operation is fast and THIS is very, very slow but then THESE operations are fast" mindset, your algorithms tend to change. It is relatively cheap to manipulate data in memory as opposed to having to fetch them from disk all the time, and it is relatively cheap to manipulate data in the cpu cache as opposed to having to fetch them from memory all the time. In other words, algorithms that take that in consideration will be faster resulting in faster results, more publications and fame and Nobel prizes.