We run 4 shards with ~ 40 GB each on the same server. An important thing you need to take care of is the use of RAM, especially in cases with large cache sizes. We allocate about 16 GB for each splinter.
You did not say anything about your qps tariffs (the number of requests per second), since processing high qps may require replication. In this case, you will store fewer fragments on the same server.
Also from our experience: the less fragments you have, the better the search efficiency for distributed interlaced search, since you spend less time merging the processor.
source share