I had to try for myself and, indeed, the ref docs are not very clear about this. In order to give the second graph a separate, independent mixed ndex
you have to adapt the properties file for the second graph like (in this case a variant on janusgraph-cql-es.properties):

gremlin.graph=org.janusgraph.core.JanusGraphFactory
storage.backend=cql

storage.hostname=127.0.0.1
storage.cql.keyspace=janusgraph2

cache.db-cache = true
cache.db-cache-clean-wait = 20
cache.db-cache-time = 180000
cache.db-cache-size = 0.25

index.search2.backend=elasticsearch
index.search2.hostname=127.0.0.1
index.search2.index-name=janusgraph2

After loading the graphs you can check against the indexing backend:

$ curl http://127.0.0.1:9200/_cat/indices
green  open .geoip_databases    O7iNY_U4Sui1yfFontk3lA 1 0 42 0 38.8mb 38.8mb
yellow open janusgraph_edges    OfbruO3BRIOPLMVYwUSWRw 1 1  6 0  4.4kb  4.4kb
yellow open janusgraph_vertices EGmMEvXuQXiIxTasEnSIDw 1 1  6 0  3.8kb  3.8kb
yellow open janusgraph2_vertices  UQITuYsnTA2d2J8MyyblLA 1 1  6 0  3.8kb  3.8kb
yellow open janusgraph2_edges     d1ZOPXVSTze0HQkg_vEz1A 1 1  6 0  4.4kb  4.4kb

Also, if you use the GraphOfTheGodsFactory for testing with two graphs, you have to use it like:
gremlin> GraphOfTheGodsFactory.load(graph1, 'search', false)
gremlin> GraphOfTheGodsFactory.load(graph2, 'search2', false)

I did not check if this can work for the ConfigurationGraphFactory, but if you find this impossible either of us should report an issue
for it (together with extending the ref docs with the above) .

Best wishes,    Marc