ORM is a by and large proven technique that has been around for a long time now (believe it or not TopLink was created in the early nineties for Smalltalk and then ported to Java). It maximizes code flexibility and database independence while minimizing boilerplate code and allowing the construction of rich domain models. Even for relatively complex domain models and queries, ORM engines tend to generate better quality SQL optimized to the underlying database than the average Java developer would with considerable time and effort. This is precisely why JPA is the persistence standard for Java EE and why a majority of modern server-side applications use ORM (indeed almost every other major development platform including .NET and Ruby on Rails positions ORM as it's default persistence solution).
That being said, there can certainly be cases where using hand-written native SQL is necessary or desirable. Most people don't seem to realize it, but JPA has always included very good support for native queries to meet these use cases (JPA 2.1 now even includes formal support for stored procedures - the ultimate form of database bound query execution). If you weren't aware of this, you should definitely read Taudo Wenzel's excellent write-up on using JPA native queries to generate reports (a common use case for native queries).
In fact you have other options for native SQL centric persistence in Java EE applications. One of the nicest ones is MyBatis-CDI. It provides a very nice CDI based abstraction for MyBatis, making it extremely easy to use in Java EE applications (in case you are wondering MyBatis is the successor to the old iBatis project). Here is how simple MyBatis CDI code can look like:
In addition to JTA transactions, MyBatis-CDI also supports local transactions. I've personally also successfully used Apache Commons DbUtils with CDI to quickly port over a legacy application with tons of hand-written SQL to Java EE/GlassFish.