HBase is useful in a lot of cases.
Lets talk scalability.
HBase can scale in linear or near linear terms. You can hash your key and get a fairly random and equal distribution across your region servers. This is critical when you consider that if you have a 10 node cluster or a 100 node cluster, pulling data (a simple get() ) will occur in a consistent time. That is to say fetching a customer record where cust_id = XXX will take the same amount of time regardless of the amount of data in the cluster.
Looking at relational databases, Oracle RAC? Anyone trying to scale that past 8-10 machines?
DB2? IBM is still improving it with stuff they inherited from Informix.
IBM was foolish to kill or rather starve Informix's XPS which also boasted near linear scalability back in the 90's.
But here's the kicker. HBase is 'free'. Although you would want to get a support contract if you were using it in a commercial production environment. So the cost of building out 100s of TB or a PB cluster is primarily the cost of the hardware.
You can't get that scalability and performance from mySQL or anything else.
Considering that HBase is < 5 years old, the technology is improving and in the last year, stability has increased.
So yeah, I'd say HBase has its place.
You just have to understand what it is and how to use it.