Everyone running a node has control over their mempool. Some node runner wants to see the whole picture and all available valid transactions so they can clearly see what's going on in the network, some wants to filter some transactions out and see the network only partly and some want to see only specific transactions in the mempool.

One noteworthy thing I noticed when playing with the filters was that, we have to download twice the transactions we try to filter. Once when we see them first time and reject it and second time when they enter in the block. It's not that bad but if one wants to optimize bandwidth usage they should not use any filters.