Marvel is indexing a set of documents for every nodes, indices and shards of your cluster. These documents are used to build the various charts.
In your case, you have many nodes / indices /shards compared to the volume of data (35 GB). That's around 100K / 80MB per index and this is not a lot.
Every 10 sec you have 11 (nodes) + 438 (indices) + 4348 (shards) = ~4800 documents indexed in 2 shards (marvel indices), so a total of 9600 docs every 10 secs which explains 960 docs /sec.
Hope this helps