think about caching strategy for datasets

[sgratzl]

In the current version all datasets are loaded on demand without any caching. However, most of the datasets could fit into the main memory.

see also https://kastnerkyle.github.io/posts/using-pytables-for-larger-than-ram-data-processing/

[ngehlenborg]

One thing to consider in this context is how to deal with situations in which many users are accessing the same server (even smallish data sets could clog main memory once there are too many that are being accessed).

Another situation is genome-wide sequencing data, which generally will not fit into main memory.